Updates

This update fixes "credstack" checks so saved provider credentials are compared against the live state your agent is actually using. What this means for you: credential slots should stay in sync after login, refresh, or failover without stale access confusing the next turn.

The /assist command now groups help scenarios with short descriptions, so finding the right command is easier. Credstack cooldown now shows as a relative countdown, and /login claude accepts a slot name so you can sign into the credential slot you intended.

WAM Agents now gives instant /login feedback, adds /cancel, and shows a background-task indicator while auth keeps working. If a prompt is too long, you get a clear context-limit reply, and after a restart your agent can recover active turns and clean stale session entries.

This update fixes /login so the "CTS code-intercept" only applies in the chat or topic where you started signing in. What this means for you: login codes from one conversation should no longer interfere with another active topic.

WAM Agents 0.123.5 is a maintenance-only release with no new commands or visible chat changes listed. What this means for you: your current sessions, access, model choices, and credential slots continue as before.

The bare /login command now manages the reserved default credential slot correctly. What this means for you: signing in without extra options updates the expected default access path instead of leaving provider credentials in the wrong place.

WAM Agents now shows credstack cooldown times on your own clock. What this means for you: when an auth slot is cooling down, the wait time matches your local time instead of a server-side timezone.

When you re-login under an existing credential slot, WAM Agents now updates that slot in place instead of leaving stale access behind. If that slot is active, your agent refreshes it immediately, so the next turn uses the credentials you just saved.

WAM Agents now detects "limit-pattern detection" cases sooner and uses "all-slots-exhausted fast-fail" when every auth slot is unavailable. What this means for you: access-limit issues should stop quickly with a clear failure instead of dragging out a stuck turn.

WAM Agents adds "credstack" for provider credentials, with slot commands, cooldown feedback, and failover that can replay the same turn when auth or usage-cap limits interrupt it. Reactions now route back to the right topic session, and "tmux" hangs get timed out so long Claude sessions do not freeze your chat.

Provider commands now support -default and -all, so you can apply model choices more broadly when switching. /status now reports Codex sessions as Codex (ChatGPT), and background-task results can still arrive after a turn finishes. Fable sessions also get their own action-style idle spinner text, while Codex preflight errors can guide an automatic re-login.

This update tightens session handling in WAM Agents: overlapping turns for the same request are locked, and stale runner sessions are cleared instead of hanging around. Dead Telegram topics now detach immediately, and wakeups re-check the latest fire time so scheduled messages land in the freshest conversation.

This update fixes reaction handling in groups: team_mode and mention_only now apply to reactions too, so your agent follows the same access rules there as in normal replies. Wakeups now attach to the freshest conversation, and Claude tmux first messages plus ExitPlanMode headings get small fixes for cleaner starts and clearer thinking topics.

This patch carries WAM Agents forward without new commands or chat-facing changes. What this means for you: your current sessions, access, model choices, and fable setup stay as they are while the update line advances.

WAM Agents now preserves your "fable" model choice when you switch modes. It also resolves the "fable" alias correctly on the TMUX path, so Claude Fable 5 stays selected instead of falling back unexpectedly.

WAM Agents now adds Claude Fable 5 support with the new /fable tier-slot. You can pick this new Claude mode from your agent when you want that reasoning profile for a conversation.

WAM Agents now reads ATX headings in ExitPlanMode and uses them in the thinking spinner. If there is no heading, it falls back to prose after ⤷, so the final status can show a clearer topic for what your agent just planned.

WAM Agents 0.120.1 is a low-key follow-up with no new commands or chat-facing fixes listed. What this means for you: your agent keeps the same sessions, access, model choices, and tmux behavior while this patch keeps the release track aligned.

WAM Agents now shows a richer thinking topic in the "⤷ slot" and carries it into the final spinner frame. What this means for you: when your agent finishes a turn, the last status gives clearer context about what it was working on.

Claude tmux mode now waits for the real end-of-turn signal before marking a session idle. What this means for you: mid-stream text should no longer make a long Claude turn look finished early.

Claude tmux mode now keeps the thinking spinner alive when an agentic turn uses tools mid-turn. What this means for you: long Claude sessions keep showing progress in chat instead of going quiet while tools run.

WAM Agents 0.119.4 is a housekeeping update with no new commands or visible chat changes. What this means for you: your existing sessions, access, and model settings continue as before while the release line stays current.

The /claude -tmux switch now completes without the old startup guard blocking it. What this means for you: changing into Claude tmux mode should feel direct, with fewer false stops before the session starts.

After /reset, Claude tmux mode now cold-starts a fresh session instead of stopping at startup. What this means for you: if you reset a conversation, your agent can bootstrap the Claude session again and continue without needing manual recovery.

Claude tmux mode now uses a per-session lock, so your agent runs one turn at a time in each session. What this means for you: overlapping Claude turns should no longer collide when a session is already busy.

Claude "tmux" mode now keeps one thinking status instead of stacking duplicate progress messages. This update also suppresses false submit errors and groups status updates per mode, so long Claude turns stay clearer in chat.

Claude tmux mode now feeds JSONL events into the thinking spinner, so you can see live progress from that path instead of waiting on a quiet session. What this means for you: long Claude turns in tmux show status while they work.

This update is a quiet version bump for WAM Agents with no new commands or user-facing fixes listed. What this means for you: your agent keeps its existing chat behavior, sessions, access, and model settings while the release track moves forward.

This update fixes Claude tmux resurrection: WAM Agents now waits for the Claude TUI prompt before handing the session back. What this means for you: after a restart or recovery, Claude turns should resume only when the session is actually ready.

Claude tmux mode now uses "bracketed paste mode" for multiline prompts. What this means for you: longer instructions with line breaks should submit as one clean prompt instead of being split or misread at session start.

Claude tmux sessions now seed `.claude.json` before startup, so the first-run setup screen no longer blocks the session from spawning. What this means for you: /claude in tmux mode should start cleanly without getting stuck on initial setup.

Claude tmux mode now checks whether its session is alive, uses auto-resurrect if it dies, and retries the turn. What this means for you: long Claude sessions are less likely to end in a dead shell after a restart or crash.

This update fixes Claude -tmux startup in WAM Agents: the startup script now runs through "/bin/sh" when direct temp-script execution is blocked. What this means for you: /claude in -tmux mode is less likely to fail right as the session starts.

WAM Agents now lets /claude start Claude in -tmux mode, with TmuxRunner keeping sessions alive and reusing them when busy. /status shows whether your current Claude session is using -tmux or -p. /login codex now stores the Codex token only, without flipping your global provider.

WAM Agents now re-checks your active reasoning provider on each turn, so model attribution stays accurate as you switch between Claude, DeepSeek, or Codex. What this means for you: the chat should show the right provider for the turn you are actually running.

This update tightens how WAM Agents verifies local tool access when a startup token is missing. What this means for you: after a restart, your agent can still authenticate its own helper processes and keep tool sessions available.

This update gives WAM Agents a longer startup grace period before access setup is treated as failed. What this means for you: after an update or restart, your agent has more time to come online instead of failing too early.

WAM Agents updates "claude-implement changes" and cleans up orphan tool processes when sessions close. The "transfer-trigger" path was fixed too, so new ownership handoff is less likely to duplicate or miss reconciliation.

This update tightens how WAM Agents starts and authenticates its tool access. What this means for you: tools come online with the right local credentials at startup, with no new commands or chat changes.

WAM Agents now handles runner errors with "400 / content-policy / 401-hint / 529-post-retry", so you get clearer next-step feedback when a turn hits policy, auth, or retry limits. Telegram media over 30 MB is now rejected before download, so oversized files fail fast.

WAM Agents adds "self-heal" recovery notices, so when something recovers after an issue, you get one clear message in your language. The update makes recovery alerts easier to understand instead of leaving you guessing what happened.

This update adds "413 fork+sanitize+replay": when a turn is too large, your agent can split, clean, and retry it instead of leaving you stuck. It also adds a "dual-token cache seam" for access checks, helping sessions keep the right credentials.

WAM Agents 0.111.1 is a small upkeep update with no new commands or visible chat changes. What this means for you: your agent keeps working with the same sessions, access, and settings while this version keeps the product current.

WAM Agents now registers /codex, /referral, /trust, and /untrust, so Codex, referral, and trust controls are available from your agent. This update also adds “safe_webfetch” protections for external content, warns on risky shell patterns, and removes deprecated Codex models.

WAM Agents now adds Claude Opus 4.8 and updates the opus alias, so you can pick the newest deep-thinking Claude model from your agent. Onboarding also gets a new mental-model step, and unhandled errors now show a friendly "/reset+/support" hint so you know the next move.

WAM Agents adds "codex-implement changes", refining Codex-based implementation turns in your agent. Quota handling now supports automatic Pro to Flash and Flash to Pro fallback, and referral details load through the right account route.

This update fixes how /runner sessions receive the project folder when your agent starts tool processes. What this means for you: file and project requests keep the right location context, so your agent can work in the intended folder instead of losing its place.

WAM Agents now passes the correct project-folder setting to your agent's tools when they start. What this means for you: file and project requests are less likely to fail because a tool launched without the right location context.

This update fixes onboarding so the Enable Replies prompt stays dismissed after you tap it. Identity edits during setup are unblocked too, so WAM Agents can prepare your agent without mistaking its own Claude files for something you cannot edit.

This update fixes path-scoped Edit/Write allow entries in WAM Agents. If you already granted access for a specific path, your agent should respect that permission instead of asking again.

WAM Agents now uses its own name consistently across user-facing text, from prompts to docs and translations. The /simple-mode spinner now shows WAM Agents naming too, keeping update text and thinking view aligned.

For quick replies, WAM Agents now auto-deletes the final thinking spinner frame after turns under 20 seconds. You still see progress while your agent works, and short chats end without an extra leftover status message.

WAM Agents 0.104.1 keeps the release track current without adding new commands or changing chat behavior. What this means for you: you can keep using your agent as usual, with the same sessions, access, and model choices intact.

WAM Agents now adapts its core instructions separately for each reasoning provider. What this means for you: when you switch models, your agent keeps the right behavior for that mode without changing how you chat.

New groups now start with "silent-by-default", so your agent waits until you explicitly bring it in instead of jumping into fresh group chats. Onboarding also adds an inline button, giving you a clearer first step when setting up WAM Agents.

WAM Agents now adds "SlashSpinner" for non-instant slash commands. When a command needs a moment, you see an animated placeholder right away, so the chat shows that your agent has accepted the command and is working on it.

WAM Agents now delivers thinking spinner updates and billing/quota CTAs silently, so your chat is not interrupted by extra notification noise. You still see the same access prompts and progress updates in place, just without unnecessary pings.

WAM Agents now uses temporary, consistent bot replies for "slash-command acks", so command confirmations don't linger in your chat. You still get the acknowledgement when a command is accepted, then the conversation stays focused on your actual work.

WAM Agents 0.99.6 is a quiet maintenance update with no new commands or visible behavior changes. What this means for you: the release line stays current while your existing chats, sessions, and agent settings keep working as before.

WAM Agents now handles reactions with the same provider as your active session, so reaction follow-ups stay tied to the right conversation. It also includes the reacted message body in context, helping your agent understand exactly what you reacted to.

WAM Agents 0.99.4 is a quiet upkeep release. What this means for you: no new commands or visible changes this time, just the current update path staying in sync.

WAM Agents now shows the Codex model indicator in the thinking spinner header. When your agent is using Codex, you can see the active model right in the chat while it works.

The /model picker now shows Codex models when you are using the Codex provider. This makes model switching clearer, so you can choose the right Codex thinking mode directly from your agent.

This update fixes Codex reasoning settings in WAM Agents: "xhigh effort" is now accepted, so your agent can use the deepest thinking mode when selected. It also removes an output-limit setting that could interfere with Codex turns.

WAM Agents now adds /login codex, giving you device-code sign-in for Codex directly from your personal agent and saving the token for later sessions. Translation access now saves in step with your stored credentials, so localized update text keeps working after setup.

WAM Agents now treats a batch of messages as one intent, with forwarded metadata plus chat and thread names included in context. In team_mode, reply_to context is expanded too, so your agent understands group replies with more of the surrounding conversation.

When your quota runs out, WAM Agents now shows a two-button "referral" option so you can invite others or continue from the right place. The same "referral" surface is now available in /limits, keeping access and top-up actions easy to find.

WAM Agents adds /referral in your personal agent, so you can open referral details right from the chat. Telegram links now keep their URLs when text is hyperlinked, mid-turn BTW messages get an hourglass reaction, and voice, photo, and doc pivots show clearer media labels while wakeups wait for pending messages.

WAM Agents now auto-deletes known slash-commands 10 seconds after you send them, keeping chats cleaner, and adds user-local time awareness through onboarding and the context block. Forum topics, reactions, team replies, identity-file edits, and reset behavior were tightened; DeepSeek now defaults to flash while preserving your per-session model choice.

This update fixes team-chat routing: when you include an @bot-mention in the message body, your agent follows that mention even if the message is also a reply. Release-note translation now uses the correct route, so update notifications arrive in the intended language.

The /team reply now shows your agent’s actual @username, so group setup instructions point to the right bot. This update also connects /team to the mention gate and fixes forum-topic auto-replies, so topic chats follow your team setting correctly.

After a crash or long downtime, your agent can now send an "I'm back" message so you know it has recovered. When quota runs out, WAM Agents now replies in Telegram with the exact capability that needs access and a clear pack purchase option.

This update is a small maintenance release for WAM Agents. What this means for you: no new commands or visible changes, just a cleaner handoff to the next version.

The new /limits command gives you one clear view of your total limits, including base access plus any packs you’ve bought. If you need more, it also shows the purchase option right there, so topping up is easier to find.

This update stops repeated subscription reminder messages from piling up in your chat. If access needs attention, you should see the notice only when it actually matters, not as spam.

This update refreshes WAM Agents' Claude reasoning layer. What this means for you: your agent stays aligned with current Claude behavior for model selection, authentication, and thinking during conversations.

This update keeps chats cleaner: when a turn finishes without an audit payload, the thinking spinner now disappears instead of lingering. You can also set "team on|off mode" for the "smart-mention gate" per chat and topic.

This hotfix turns the "smart-mention gate" off by default. In group chats, your agent should no longer stay silent because that gate misread whether a message was meant for it.

WAM Agents now uses "smart group mention" handling, so group replies follow the message text first. This update also adds "per-message capability gating": when a specific capability needs more quota or access, you see upgrade text for that exact action.

Your agent no longer calls /v1/models when you use Claude OAuth, avoiding an unnecessary model check during authentication. BYOK Claude sessions now bypass DeepSeek-token quota checks, so your personal Claude key is not blocked by the wrong quota counter.

When your quota runs out, WAM Agents now sends you straight to @wamagentsbot to buy a new pack. This keeps the next step clear, so you can restore access without searching for where to top up.

This update fixes how your own API key is handed to your agent’s tools after it loads from the secure vault. What this means for you: requests that depend on your personal key keep access during reasoning instead of failing because the key was missing from the session.

WAM Agents 0.88.3 is now live. This release focuses on stability and configuration reliability — your agent now gracefully handles missing billing endpoint settings, ensuring quota checks and /status information stay accessible across all deployment setups without interruption.

WAM Agents 0.88.2 now accepts the BILLING_API_URL environment variable as a fallback for your billing endpoint configuration, so setup stays flexible whether you're using config files or environment variables. This keeps quota checks and /status information reliable across different deployment setups without configuration gaps.

Your agent now authenticates correctly with the billing system when checking your token limits — the Authorization Bearer header is now always sent to the validation endpoint, so /status displays accurate quota information without gaps.

WAM Agents 0.88.0 now routes inactive-subscription users to @wamagentsbot so you stay connected to the right account. We've also refined how your agent detects missing provider secrets—it now logs at INFO level instead of flagging a warning when you're in vault-mode, keeping your startup logs cleaner without false alerts about credentials that are actually secure.

WAM Agents 0.87.0 now skips the restart notification by default, so updates roll through silently without interrupting your conversations — you can still see the changelog anytime by checking your chat history if you want to know what changed.

WAM Agents 0.86.0 now discovers your billing provider secrets dynamically instead of using hardcoded keys, keeping authentication flexible as your infrastructure evolves. We've also tightened how your agent writes secrets to the vault—closing three gaps in the credential sync process so your authentication tokens stay in sync across storage layers without loss or duplication.

Your agent now loads knowledge more leanly at startup — the pre-baked knowledge layer has been removed, so your agent fetches your custom context dynamically instead of carrying bundled files. This keeps initialization faster and ensures your agent always has the latest version of your identity and knowledge without stale copies slowing things down.

WAM Agents 0.84.10 now ensures your /login command closes all active reasoning sessions when you authenticate, keeping your model selection in sync across all your conversations instead of leaving stray runner processes open. This means re-authenticating your Claude credentials stays clean and seamless without orphaned sessions interfering with your chats.

WAM Agents 0.84.9 refines how your /login command works across conversations — when you authenticate your Claude credentials, your agent now sweeps all active sessions to keep your model selection in sync, while preserving your choice so you don't lose your preferred reasoning provider mid-authentication. This ensures your provider settings stay consistent across all your chats without unexpected reverts when you're re-authenticating or switching credentials.

Your agent now resets its reasoning model back to Claude when you complete the /login command, so after authenticating your credentials, your agent returns to your default provider instead of carrying over a previous selection. This ensures fresh starts feel consistent and predictable when you re-authenticate.

Your agent now ensures your Claude authentication token stays in sync across your vault and environment—when you log in or switch reasoning providers, your credentials write cleanly to both storage layers without gaps. If your vault secret ever becomes empty, your agent automatically falls back to your environment variable, so authentication stays reliable and uninterrupted across restarts and token refreshes.

WAM Agents 0.84.6 now ensures your Claude authentication token stays in sync across your vault and environment — when you log in or switch reasoning providers, your credentials write cleanly to both storage layers without gaps. If your vault secret ever becomes empty, your agent automatically falls back to your environment variable, so authentication stays reliable and uninterrupted across restarts and token refreshes.

Your agent now downloads your identity files as individual content blobs and assembles them into a complete bundle at startup — fixing read-only file conflicts so your custom context loads reliably every time without interruption.

Your agent now properly copies knowledge files to writable locations during startup, so read-only file conflicts no longer block access to your custom context. This ensures your agent loads your identity and knowledge reliably without interruption, keeping your conversations seamless across restarts.

Your agent now properly copies knowledge files to writable locations during startup, so read-only file conflicts no longer block access to your custom context. This ensures your agent loads your identity and knowledge reliably without interruption, keeping your conversations seamless across restarts.

Your agent now loads your identity and knowledge more reliably at startup — the knowledge file structure is flattened for cleaner access, and identity verification now uses content hashing instead of file paths, so your custom context arrives complete without gaps across restarts.

WAM Agents 0.84.1 refines how your agent verifies your identity and knowledge at startup—client signature checking now matches the canonical union format, so your custom context loads reliably without verification gaps. This ensures your agent has complete access to who you are and what you know the moment it boots up, keeping your conversations seamless across restarts.

WAM Agents 0.84.0 refines your knowledge and identity publishing pipeline with improved CI workflows and Tailscale routing—your agent now fetches your identity and knowledge more reliably at startup and during scheduled hourly refreshes. Behind the scenes, the migration canary E2E gate validates new releases before wider rollout, and the signing key infrastructure is updated to ensure secure credential handling, so your agent's startup stays smooth and your custom context loads without interruption.

Your agent now correctly resolves secrets from your vault socket when it boots up, fixing a critical issue where missing container connections were causing authentication failures. We've also refined how your agent reads credentials from your environment — it now prioritizes your .secrets/.env file over Docker variables, ensuring sensitive data flows through the right channels without configuration gaps that could interrupt reasoning or tool access mid-turn.

WAM Agents 0.83.0 now validates direct API calls against proxy routing with a shadow comparator, ensuring your reasoning, speech-to-text, web search, and image analysis stay aligned during the migration to unified proxy systems. We've also added vault HTTP client support so your agent reads secrets through a secure resolution chain, keeping sensitive credentials accessible and protected without configuration gaps or fallback delays.

Your agent now includes the correct proxy address in its subprocess environment during startup, fixing a critical issue where backend connections were failing silently. This ensures all your agent's capabilities stay connected and responsive without routing gaps that could interrupt reasoning or tool execution mid-turn.

WAM Agents 0.82.3 now prevents your agent from misinterpreting environment variable templates during MCP initialization — literal ${VAR} patterns no longer get expanded incorrectly, keeping your credentials and secrets safe from accidental substitution. The /reset command also now properly clears your reasoning model selection, so after a reset your agent returns to its default provider instead of carrying over a previous choice.

WAM Agents 0.82.2 fixes image handling when you're using reasoning providers that don't support vision — your agent now skips image processing hints for those models, keeping reasoning flowing smoothly without capability mismatches. This ensures you can switch between Claude and non-vision providers without unexpected processing errors when images are involved in your conversations.

WAM Agents 0.82.1 now ensures your reasoning broker, speech-to-text, web search, and image analysis all receive the correct authentication credentials during startup—fixing a critical issue where environment variables weren't being passed to the MCP subprocess. This means all your agent's capabilities stay connected and responsive without credential gaps that could interrupt reasoning or tool execution mid-turn.

WAM Agents 0.82.0 now routes your reasoning, speech-to-text, web search, and image analysis through a unified proxy system — ensuring consistent performance and security across all your agent's capabilities. Behind the scenes, we've added a shadow comparator so direct API calls and proxy routing stay aligned, and your agent now reads secrets from vault as its primary source, keeping sensitive credentials secure without fallback delays.

Your agent now pulls your identity and knowledge at startup with automatic hourly refreshes, giving it complete context about who you are and what you know without manual setup. Secret management is also simplified — your agent reads credentials from a union vault socket with a plaintext .env fallback, so sensitive data stays secure and accessible during reasoning without extra configuration steps.

WAM Agents 0.80.0 now keeps your reasoning broker alive during idle periods with better heartbeat monitoring, so your agent stays responsive even when you're not actively chatting. We've also added structured exit logging for tool processes, making it clearer when something goes wrong with MCP server connections — troubleshooting reasoning failures becomes faster because you'll see exactly where the tool chain stumbled instead of silent dropouts.

WAM Agents 0.79.1 fixes two critical issues from the previous release. Tool call errors in your SSH and server utilities no longer crash the reasoning broker — bad requests now fail cleanly without taking down the entire MCP process. We've also resolved a provider-selection bug where switching from Claude to DeepSeek with /deepseek could leave your agent using stale credentials, causing reasoning to fail mid-turn — your agent now correctly flushes the old provider state when you change reasoning engines, so model switching stays reliable every time.

WAM Agents 0.79.0 now routes your /status quota query through the reasoning broker when you're on the DeepSeek or broker tier, giving you accurate token consumption data no matter which reasoning provider you're using. This means checking your usage stays consistent across Claude and DeepSeek without authentication gaps or mismatched billing information.

WAM Agents 0.78.3 fixes a critical issue where image analysis was failing during reasoning — the image_file field has been corrected to image, so your agent can now process photos and visual content without errors when you ask it to analyze images mid-conversation.

Your agent now has access to image analysis and web search during reasoning—when your agent tackles requests that need visual details or live information, it can pull both capabilities without extra setup. These tools round out your agent's default toolkit so complex conversations flow naturally without permission gaps.

WAM Agents 0.78.1 fixes a critical issue where your agent could lose connection to the reasoning broker during startup — it now correctly falls back to the platform broker endpoint if the primary connection fails, keeping your reasoning seamless across restarts without interruption.

WAM Agents 0.78.0 now lets you apply the mention setting across all your group chats at once with /mention on -group, so you can lock down your agent's responses to direct mentions everywhere without toggling each chat individually. We've also expanded your agent's reasoning toolset with web_search and analyze_image capabilities, giving it richer ability to pull live information and process visual content during conversations. When Claude Code becomes unavailable, your agent now detects the contract violation and retries automatically instead of getting stuck, keeping reasoning flowing smoothly even during brief interruptions.

WAM Agents 0.77.0 fixes a critical routing issue where your reasoning provider choice was getting mixed up between conversations — after you complete a turn, your agent now correctly saves which model you selected so switching between Claude and DeepSeek stays reliable across topics without unexpected reverts. This closes a gap where your provider preference could drift mid-session, keeping your reasoning seamless and accurate every time you chat.

Your agent's reasoning spinner now displays a consistent lightning bolt emoji as fallback when custom provider symbols aren't available, fixing a visual glitch where the spinner header could become unreadable during reasoning turns. This keeps your display clean and stable across all your conversations, whether you're on Claude or DeepSeek.

WAM Agents 0.76.1 removes custom emoji from your reasoning spinner header, keeping the display clean and consistent across all your conversations. This fix closes a visual gap where provider symbols could clutter the spinner during reasoning turns — your agent now shows a streamlined header that stays readable whether you're on Claude or DeepSeek.

Your agent now syncs with Claude's latest reasoning implementation, ensuring model selection and provider routing stay reliable across all your conversations. This keeps your reasoning seamless and accurate when you switch between providers — giving you stable, consistent performance without timing gaps or configuration drift.

WAM Agents 0.75.3 fixes two issues from the previous release. Guest access control in channels now respects the mention_only setting properly, so your agent won't respond in locked topics unless you explicitly mention it—matching the behavior for channel owners. We've also routed cohort notifications through the wamalert system, ensuring you stay informed about staged rollouts and feature updates without missed alerts.

WAM Agents 0.75.2 fixes three reliability issues from the previous release. The /claude authentication now correctly sets your active reasoning provider across all your conversations instead of just the current one, so switching to Claude stays persistent without unexpected reverts. Your agent also now displays the correct model name from its bundled settings instead of outdated cached versions, and custom provider emoji remain visible in the reasoning spinner header throughout your entire turn. These fixes close gaps where model selection could drift between topics and display information could fall out of sync mid-conversation.

WAM Agents 0.75.1 tightens security around billing checks — your agent now guards quota-status calls against permission errors when running in client containers, so billing lookups stay reliable without interruption. We've also fixed the transfer-trigger permission guard to prevent crashes when your agent boots in restricted environments, and resolved an issue where custom provider emoji weren't visible to other users in group chats. The /login scope handling is now corrected so your token stays saved globally but your active reasoning provider selection no longer flips unexpectedly between topics — each conversation respects your model choice without drift.

WAM Agents 0.75.0 fixes a critical issue where Claude reasoning was returning empty results — your agent now properly captures tool outputs and reasoning feedback instead of dropping them silently mid-turn. We've also streamlined how your agent stores OAuth credentials, moving them to secure container storage so billing checks stay clean and separate from authentication. Model selection now carries over correctly after you switch providers with /provider, ensuring your first turn uses the right reasoning engine instead of reverting to a previous choice.

WAM Agents 0.74.1 fixes how your agent authenticates with the billing system — the X-Client-Hash header is now always included when checking token limits, so you'll see accurate /status information without authentication gaps. This release also ensures reasoning output stays cleanly separated from your chat replies, keeping conversations crisp and focused without stale results bleeding through.

WAM Agents 0.74.0 now cleanly separates reasoning output from your chat replies — leftover results from previous turns no longer bleed into new messages, giving you crisp, focused conversations without stale reasoning leaking through. Your agent also injects your bot username directly into its Telegram context so it understands how to reference itself and handle mentions correctly. Behind the scenes, the fallback to standard output is now smarter about when Claude Code is actually unavailable, and the busy-wait blocker aligns with the latest Claude Code schema, so your agent reasons reliably without getting stuck in resource-draining loops.

WAM Agents 0.73.0 refines how your agent routes reasoning across providers — the broker now switches provider environments based on your tier instead of leaking platform secrets into subprocesses, keeping sensitive credentials secure. This release also syncs with Claude's latest implementation layer to keep your model switching and reasoning precise across all conversations.

Your agent's reasoning spinner now displays the actual cost of your current turn using per-model pricing — so you'll see exactly what Claude or DeepSeek reasoning costs as it runs, with accurate numbers for whichever provider you've chosen. This release also syncs with the latest Claude implementation layer to keep your model switching and reasoning precise across all conversations.

WAM Agents 0.71.0 refines how your agent syncs with Claude's reasoning layer, ensuring model selection and provider routing stay reliable across all your conversations. Behind the scenes, the implementation layer is streamlined so environment setup is cleaner and reasoning stays precise when you switch between providers — giving you seamless, accurate conversations without timing gaps or configuration drift.

WAM Agents 0.70.0 cleans up how your agent loads reasoning provider settings — the legacy environment fallback is gone, so model configuration stays reliable and direct without redundant code paths. The reasoning spinner header is also rebuilt to display your active model's proper name, and the /model picker now syncs perfectly with what's actually running, so switching reasoning depth feels seamless and accurate across all your conversations.

WAM Agents 0.69.0 brings a cleaner command and capability system — your agent now reads commands and tools from a unified registry, so what you can access and what your agent can do stay perfectly in sync without drift between systems. Behind the scenes, the reasoning layer is refactored for stability and the /assist help menu is now streamlined to pull directly from the command source, cutting out redundant copies that could fall out of step. This release completes the migration to a modern agent architecture, making future updates faster and more reliable.

WAM Agents 0.68.0 now consolidates your reasoning model commands into a unified /model interface — the legacy /opus, /sonnet, and /haiku shortcuts still work, but /model is now your primary way to switch reasoning depth across all conversations. The commands registry is also unified with your capabilities list, so what you can access and what your agent can do stay perfectly in sync without drift between systems. This release also syncs with Claude's latest implementation layer to keep reasoning stable and precise as you work.

WAM Agents 0.67.0 refines how your agent syncs with Claude's latest implementation layer, ensuring reasoning stays reliable across all your conversations. Behind the scenes, environment setup is cleaner and provider routing more precise, so model switching between Claude and DeepSeek continues working smoothly without timing gaps or permission misroutes.

WAM Agents 0.66.0 refines how your agent routes conversations across different reasoning providers — environment variables are now properly set when your agent starts, fixing a timing issue where permissions could get misdirected between Claude and DeepSeek sessions. The reasoning spinner also correctly displays which backend is active when you're on the DeepSeek tier, and forum topics now support the /mention command so you can grant per-topic access control in group chats just like direct conversations.

WAM Agents 0.65.1 now routes support requests through the /support command when you mention @wamsupport, preventing your agent from fabricating answers to questions it can't handle. This ensures you get genuine help instead of hallucinated responses when you need real assistance.

WAM Agents 0.65.0 now fetches your reasoning model's context window dynamically based on which provider you're using — Claude and DeepSeek each get the right capacity limits without hardcoded guesses, so your agent accurately tracks how much conversation history fits in each turn. This means model switching stays precise and you'll never hit unexpected capacity warnings that don't match your actual reasoning engine.

Your agent now syncs its reasoning layer with Claude implementation updates, ensuring model selection and authentication work together smoothly across conversations. This means switching between reasoning providers stays seamless without legacy session data interfering, so your agent picks the right engine every time you chat.

WAM Agents 0.63.0 brings unified reasoning provider handling — your agent now correctly maintains which model you chose across sessions, so switching between Claude and DeepSeek stays reliable without legacy session data interfering. Behind the scenes, the implementation layer for Claude reasoning is refined to ensure authentication and model selection work smoothly together, giving you seamless provider switching without timeouts or mixed reasoning engines mid-conversation.

Your agent now handles reasoning provider switching more reliably—legacy session data no longer blocks authentication during the brief window when cache refreshes, so model changes between Claude and DeepSeek complete smoothly without timeout interruptions.

Your agent now respects which reasoning model you chose in each conversation — switching between Claude and DeepSeek with /claude or /deepseek, then checking /status, no longer shows a mismatched provider. This fix closes a gap where your agent could report the wrong reasoning engine for a session, so you'll always see accurate model info without confusion.

WAM Agents 0.62.1 now prevents your agent from accidentally mixing reasoning providers when you switch to DeepSeek — legacy session data no longer bleeds across Claude and DeepSeek contexts, so model switching stays clean without stale provider state interfering with your conversations.

WAM Agents 0.62.0 now ensures your reasoning model choice persists correctly across sessions — when you switch to Claude with /login or select a model with /claude, your preference stays locked in for that conversation without mixing providers mid-turn. This closes a gap where scheduled automations could lose track of which reasoning engine they were using, so your agent now picks up exactly where it left off when it wakes up at a scheduled time.

Your agent now correctly routes scheduled wake-ups to the right reasoning model—previous versions were losing provider context and inflating session keys, causing your agent to miss model assignments when it woke up at scheduled times. This fix ensures automations pick up exactly where they left off with the right reasoning engine, no more context gaps.

New agents now start without preset names or model labels, so your first agent arrives as a blank canvas ready for you to shape. This simplifies the initial setup experience—you can immediately customize your agent's identity and reasoning depth without inheriting default branding.

Your agent now switches to DeepSeek automatically when you log out from Claude, so you stay connected to reasoning without interruption. Model selection settings are also saved more reliably—the system now writes your preference atomically, so configuration changes stick cleanly even during heavy load.

WAM Agents 0.60.1 improves stability during updates—your agent now runs end-to-end rollout tests asynchronously so new versions validate reliably without blocking your conversations. Behind the scenes, legacy session slots now migrate cleanly when your agent switches reasoning providers, ensuring you never lose context when moving between Claude and DeepSeek.

WAM Agents 0.60.0 removes the permission gate for DeepSeek reasoning — you can now switch to DeepSeek without a one-time consent prompt, giving you faster access to your preferred model. The /login command now automatically sets Claude as your active reasoning provider after you authenticate, and confirmation messages now start with a checkmark so you spot them instantly in your chat. Behind the scenes, the authentication token rotation is now working reliably, and /logout properly clears your Claude credentials from memory so stale tokens don't interfere with future sessions.

WAM Agents 0.59.0 now routes your reasoning through a unified provider registry, so model switching with /claude and /deepseek works consistently across all your conversations. Behind the scenes, your agent now tracks token consumption per turn with complete accuracy — you'll see exactly how many tokens each message costs in your /status breakdown instead of approximations.

WAM Agents 0.58.1 fixes a timezone handling issue that was causing the grandfathered user check to fail unexpectedly — your agent now correctly identifies whether you have early-access privileges regardless of how the system records timestamps internally.

Your agent now shows live progress feedback when you use /login—the reasoning spinner updates in real time as authentication completes, mirroring the same responsive experience you get across other commands. This means you'll see exactly what's happening during credential setup instead of waiting silently for the process to finish.

The /status command now displays your actual token consumption using the correct billing fields — you'll see your used tokens and total tokens at a glance instead of a calculated sum that sometimes didn't match your real usage. This gives you clearer, more accurate visibility into what you've spent in your current period.

Your agent now filters uploaded documents by type — text files pass through cleanly, but binary media is blocked to prevent processing friction. We've also fixed the billing proxy so it properly authenticates with your credentials when reaching the payment system, ensuring access checks work seamlessly without connection errors.

WAM Agents 0.56.0 now routes your /deepseek reasoning through a consent gate—new users see a one-time permission prompt before switching to DeepSeek, while grandfathered users access it freely. The /model command gains flash and pro shortcuts for faster model selection, with flash set as your default, and the /assist menu is now more compact with an inline-keyboard drill-down so help text stays clean and scannable. The /claude command now auto-authenticates when you have no OAuth token, or switches models instantly if you're already logged in—no extra steps needed to change reasoning depth.

WAM Agents 0.55.0 now syncs your reasoning model selection across all interfaces — the /model command and provider aliases like /opus, /sonnet, and /haiku are now unified with the Claude implementation layer, so switching reasoning depth works consistently everywhere you chat. The reasoning spinner also displays which model is actively processing your request, giving you clear visibility into whether Claude or DeepSeek is handling your turn.

Your agent now shows /claude, /deepseek, /status, and /logout in the /assist help menu so you can easily discover model-switching and session commands. We've also fixed a model name typo in the deepseek haiku option, so switching to that reasoning depth now works without errors.

WAM Agents 0.54.0 now syncs your reasoning model configuration with the Claude implementation layer, ensuring commands like /claude and /deepseek stay aligned with how your agent actually processes requests. Behind the scenes, we've split billing logic into cleaner provider-state and provider-environment modules, and integrated dual-session routing so your agent handles Telegram messages and Claude reasoning in lockstep without coordination gaps. The continuous integration pipeline also now runs rollout tests on the correct agent infrastructure, so stability checks catch real issues before updates reach you.

WAM Agents 0.53.0 now gates new releases through a stability test stage before wider rollout — your agent validates key features in a controlled environment first, so updates reach you only after they pass real-world checks. We've also improved the continuous integration pipeline to properly handle issue tracking, making the release process more reliable behind the scenes.

WAM Agents 0.52.0 now skips duplicate changelog summaries when your agent updates across multiple versions—you'll see only the latest release notes instead of stacked notifications, keeping restart updates clean and focused. We've also ensured your agent loads complete product context by default, so it has full awareness of its capabilities and your setup from the moment it starts.

Your agent now displays what direction you most recently steered it in the reasoning spinner footer during interactive reasoning — letting you track your pivots at a glance. We've also fixed a timing issue where the pivot counter wasn't resetting cleanly after edits, keeping the spinner accurate when you guide your agent mid-turn in /runner mode.

WAM Agents now displays your last pivot preview in the reasoning spinner footer, so you can track which direction your agent most recently shifted during interactive reasoning. We've also fixed a timing issue where the inject counter wasn't resetting cleanly after a result event with force_edit, ensuring the spinner stays accurate when you steer your agent mid-turn in /runner mode.

When you pivot your agent mid-turn in /runner mode, the reasoning spinner now updates instantly instead of waiting — giving you real-time feedback as your agent shifts direction during interactive reasoning.

Your agent now handles mid-turn message injection more reliably during interactive reasoning — when you send a message while your agent is thinking in /runner mode, it processes your input without blocking the live-thinking stream, making conversations feel seamless and responsive.

Your agent now logs MCP errors during interactive reasoning so you can spot tool failures when running in /runner mode — making troubleshooting smoother when live-thinking conversations need debugging. We've also added a per-envelope reminder for the MCP system, so your agent stays aware of available tools throughout extended interactive sessions.

WAM Agents 0.48.1 now lists the /runner command in your /assist help text so you can easily discover interactive reasoning mode. We've also refined how your agent passes reinforcement signals during live-thinking sessions, making interactive conversations smoother when you toggle between standard and runner modes.

Your agent now has a /runner command to toggle interactive reasoning mode globally, letting you switch between standard and live-thinking conversations on demand. Behind the scenes, media batching is now smarter — your agent waits 700 milliseconds before grouping photos and voice messages, catching all items in a quick burst so albums form naturally without false splits. These changes lay groundwork for deeper interactive features coming soon.

WAM Agents 0.47.0 brings smarter changelog summaries — your agent now generates concise, tone-matched update notes and translates them per language with literal-citation accuracy, so you see exactly what changed without fluff. Behind the scenes, the spinner no longer blocks on long turns, letting you interact with your agent while it's thinking deeply.

Your agent now automatically transfers to your ownership the moment you send the bot its first message. The transfer used to require manual Telegram confirmations and would sometimes hang for 7 days — now it happens instantly, with no extra steps from you.

The /stop and /reset commands now cleanly close your session — media files from a recent album no longer leak into already-cancelled conversations. The update process is also stabilized: new versions properly wait for container readiness and no longer crash on invalid forum-topic identifiers.

WAM Agents 0.45.0 now rolls out updates in stages — your agent automatically handles stability checks before moving to wider release, and you can manually trigger beta or full rollouts when you're ready. This means new features reach you faster while keeping your conversations stable and predictable.

The Voice, Photo, and Command preset buttons have been removed from your agent's initial greeting — you now land directly in a normal conversation, skipping the format-choice step. The change simplifies onboarding so you can start talking to your agent right away without intermediate prompts.

WAM Agents 0.44.0 fixes a crash in the wamalert worker that was causing monitoring alerts to fail silently — your agent now stays reliably connected to system notifications without unexpected interruptions.

WAM Agents 0.43.1 locks Claude Code to version 2.1.119, pulling in upstream fixes and ensuring your agent's reasoning runs consistently across restarts without unexpected behavior changes.

WAM Agents now batches media more reliably—photos and voice messages you send in quick succession are grouped by their media_group_id, so your agent receives them as a cohesive album rather than scattered items. New group chats also default to open mode, letting your agent respond freely without requiring a mention unless you lock it down later.

WAM Agents 0.42.0 now passes /clear directly through to Claude Code, so you can wipe reasoning state mid-conversation without losing your chat history. First-run setup now includes an inline keyboard for Voice, Photo, and Command options right in the greeting, and your agent ships with a baseline .claude/settings.json template so custom reasoning configs have a clear starting point. We've also separated billing timeouts from reply timeouts with smarter retry logic, and fixed a language-switching regression in the reasoning spinner so status updates respect your Telegram language setting consistently.

Your agent now supports per-topic guest access control — you can grant specific guests read or write permissions within individual forum topics without affecting other conversations. This means you can safely collaborate in group chats and forums with fine-grained privacy, keeping sensitive discussions locked while opening up public topics to trusted participants.

WAM Agents 0.40.1 fixes a critical issue where the /compact and /context commands weren't being properly passed through to Claude Code during reasoning — your agent now respects these settings when controlling context carryover and summary compression, so conversation continuity works as intended without command loss during processing.

WAM Agents 0.40.0 adds /compact and /context commands so you can tailor how much context your agent carries between messages — compress old reasoning into summaries or pull full conversation history when you need continuity. We've also fixed a timing gap in the reasoning spinner where the typing indicator would sometimes drop out during long reasoning turns, keeping your chat feeling responsive even on complex requests.

WAM Agents 0.39.0 now stops your agent from getting stuck in busy-wait loops during tool execution — the PreToolUse hook blocks runaway Bash processes before they drain your resources. We've also synced your project card workflow so it automatically marks tasks as Status=Done when you close an issue or merge a pull request, keeping your project board in step with your actual work without manual updates.

WAM Agents 0.38.1 fixes a critical issue where the GitHub CLI extension installation was failing due to outdated token permissions — the agent now uses classic-scope credentials for this step, so your agent can properly set up GitHub operations without authentication errors during initialization.

Your agent now handles media batches more cleanly—false queue warnings and stale prefixes no longer appear when you send multiple photos or voice messages in quick succession. Stale topic sessions are also automatically reconciled, and the chat-action sender now recovers gracefully from Telegram rate limits and network hiccups, so your agent stays responsive even during heavy load or connectivity issues.

The /stop command now cancels any pending permission prompts and reports how many were waiting, so you can cleanly exit conversations without leaving stray access requests hanging. Your agent also cleans up orphan media files when you stop a session and runs background garbage collection, keeping storage tidy without manual intervention. We've also simplified the inbox system by removing retry limits and the mark-failed logic, so message handling is more straightforward and predictable.

Your agent now has per-thread /effort command control, letting you adjust reasoning depth individually within each conversation topic — set one chat to quick responses and another to deep thinking without affecting your global settings.

WAM Agents now displays how many requests are queued ahead of your current message during long turns — you'll see a rotation symbol with the count and wait time in your reasoning spinner, so you know exactly how busy your agent is without guessing when your turn will come.

WAM Agents 0.35.0 now displays your scheduled task wake-up time directly in the reasoning spinner, always shown in Moscow Standard Time so you can track exactly when your agent will next run on its own. This means you'll have complete visibility into your automation timeline without needing to calculate time zones or guess when background workflows will trigger.

WAM Agents 0.34.0 now verifies your bot access through billing lookups instead of manual token checks—your agent confirms its permissions instantly and securely, so there's no delay when you start a conversation or switch between chats.

WAM Agents 0.33.2 now cleans up duplicate OAuth credentials during startup — if your credentials.json already has a token, the agent drops the CLAUDE_CODE_OAUTH_TOKEN environment variable so conflicting authentication sources don't interfere with your reasoning sessions.

WAM Agents 0.33.1 fixes scheduled task execution — your agent no longer re-runs completed tool calls when it wakes up at a scheduled time, so automation workflows execute cleanly without duplicating previous work.

WAM Agents 0.33.0 now includes scheduled task execution, letting your agent wake up and run reasoning at specific times without you needing to send a message — perfect for automations, reminders, and time-based workflows that work while you're away.

WAM Agents 0.32.1 now handles network hiccups more gracefully — when Telegram rate-limits your agent or the connection drops, it automatically falls back to standard output and restarts the reasoning spinner without losing your place in the conversation.

WAM Agents 0.32.0 now lets you authenticate and grant access on a per-chat basis using /mention — your agent can now manage permissions directly within specific conversations without global settings, giving you finer control over what each chat can access.

WAM Agents 0.31.3 improves reliability during agent restarts — it now correctly identifies out-of-memory shutdowns as retryable events instead of permanent failures, so your agent can recover and pick up where it left off. We've also fixed a critical issue where stale session identifiers could persist after Claude updates, causing your agent to lose context with Telegram — your agent now detects and clears this state automatically, ensuring seamless continuity across restarts.

WAM Agents 0.31.2 fixes several environment and storage issues from the previous release. Your agent now properly restores its working directory when session context is missing, persists your /allow permissions to a writable location instead of a read-only system folder, and correctly passes your bot token and SSH credentials to Claude during reasoning — so all your configured access and tools work as expected without interruption.

WAM Agents 0.31.1 fixes credential handling during Claude reasoning — your agent now properly loads secrets from your environment and filters them safely when spawning reasoning processes, so sensitive data stays protected without blocking access to the tools that need it.

WAM Agents 0.31.0 expands end-to-end testing to cover forum topic dispatch, agent identity persistence across restarts, document uploads, and reaction handling — catching real-world issues before they reach you. We've also fixed a timing race in the billing system during agent startup and improved how release notes are translated and published, so you get clearer, faster updates when new versions roll out.

WAM Agents 0.30.1 tightens how usage limits are detected during reasoning, so false alerts about hitting your cap no longer interrupt conversations — you'll only see warnings when you're actually approaching a real threshold.

Your agent's reasoning spinner now displays tool names in plain language instead of technical identifiers — so when it calls a capability during thinking, you'll see something like "fetch web page" instead of an internal code reference, making it clearer what your agent is actually doing as it reasons through your request.

WAM Agents 0.29.0 refines your reasoning spinner with richer visual signposting — you'll now see active execution paths, semantic topic headers, skill invocations marked with command symbols, scheduled task timing, your current worktree branch, memory writes logged inline, and cleaner thinking extraction. These changes make it easier to track what your agent is doing during complex reasoning without cluttering the display.

WAM Agents 0.28.0 streamlines your agent setup by renaming the identity file from soul.md to agent.md and reorganizing the onboarding flow so you get to work faster. Secret management is also overhauled — your agent now reads credentials from a plaintext .env file instead of the older MCP tool system, giving you simpler, more direct control over what sensitive data your agent can access during conversations.

WAM Agents 0.27.0 now ships its core system layer as a mounted volume, making agent initialization faster and more reliable across restarts. We've also added end-to-end black-box testing on live deployed agents, so every release is verified against real-world conversation flows before it reaches you — this means fewer edge cases slip through and your agent stays stable as features roll out.

WAM Agents 0.26.0 now bridges your agent to GitHub operations via SSH — your agent can query repositories, check workflows, and pull code details directly when you need it. We've also bundled GitHub CLI into the agent image, so remote Git tasks execute without extra setup delays.

WAM Agents 0.25.0 now loads your environment secrets automatically when your agent runs Claude reasoning, so sensitive credentials stay secure without manual setup. You can also fine-tune SSH access per host with a custom whitelist — your agent can now use read-only tools via SSH when you grant permission, giving you controlled remote access without exposing write commands or system changes.

WAM Agents 0.24.2 fixes a packaging issue that was breaking release notes during updates — the build system now includes all required dependencies, so you'll see clean, complete changelog messages when your agent restarts instead of silent failures behind the scenes.

WAM Agents 0.24.1 fixes a critical issue where the system prompt was appending duplicate context to your agent — this was breaking mentions and references in your identity file. The update removes that redundant append step, so your agent now reads your identity cleanly without losing @ mentions or disrupting how it understands your contacts and available tools.

Your agent now has secure remote access as an opt-in tool through SSH — you can grant it controlled command-line permissions when you need it to interact with your systems. We've also ensured the default agent personality now includes all your identity references built in, so your agent has complete context about your soul, contacts, and available tools from the very start.

WAM Agents 0.23.0 adds the /force command so you can immediately terminate all running sessions when you need an update right away, rather than waiting for natural completion. We've also fixed two storage issues: topic names and project graphs now save to a writable temp directory instead of read-only locations, and duplicate message thread identifiers in group chats no longer cause crashes when the agent falls back to standard output logging.

WAM Agents now automatically reconciles your topics every 24 hours, so deleted forums are caught and cleaned up without manual intervention. Token usage is now logged per response into your session record, giving you a detailed breakdown of what each message costs. The agent also supports persistent delegation loops with project graphs, letting you build multi-step workflows that survive restarts. A public changelog page is now live at agents.magnets.tg/changelog for easy version tracking.

WAM Agents 0.21.0 refines how your agent handles authentication token expiry — it now recovers more gracefully when credentials need refresh, and retry countdowns give you clearer feedback with smarter backoff timing. Behind the scenes, session counters now track only conversations with active reasoning, so you get an accurate picture of what your agent is actually processing.

The onboarding introduction now appears only in direct messages with your agent, not in group chats — this keeps setup guidance focused and prevents clutter when you mention your agent in a topic or channel.

When your agent updates across multiple versions, you now see only the latest summary instead of stacked changelogs — keeping restart notifications clean and focused on what's actually new in this release.

WAM Agents 0.20.0 now presents your changelog in a storytelling format when the bot updates, making release notes feel like a natural part of your conversation rather than a technical list. This means you'll understand what changed in plain language, focused on how each update actually affects your work with your agent.

WAM Agents now shows you the changelog in your own language when it updates, with unnecessary links removed so the restart notification stays clean and focused on what actually changed for you.

WAM Agents now displays your changelog in your language when the bot updates, with GitHub links stripped out so you see clean, readable release notes. This means restart notifications are fully localized and focused on what matters to you — no unnecessary formatting or external references cluttering the update message.

WAM Agents now properly saves your session state during restarts, so when your agent shuts down and comes back online, it retains context about what was running — no more lost progress or orphaned conversations when updates happen.

WAM Agents 0.18.2 ensures the changelog is always bundled with your agent during updates, so you'll never miss release notes when the bot restarts. This fix closes a packaging gap from the previous version — your agent now has complete version history built in every time.

WAM Agents 0.18.1 now bundles the version and changelog directly into the agent container, so you'll always see the exact release notes when your agent updates — no more missing context about what changed between restarts.

WAM Agents 0.18.0 now shows you exactly what changed between versions when your agent restarts — you'll see a detailed diff of the changelog so you know what's new without having to hunt for it yourself.

WAM Agents 0.17.0 now speaks your language—system messages, error alerts, and the Plan header in the reasoning spinner are all localized to match your Telegram settings. You'll also get clearer, human-readable error messages when something goes wrong with data processing, making it easier to spot what needs fixing. This release refines shutdown behavior so your agent drains sessions cleanly and respects your simple-mode preference from the moment you create it.

WAM Agents now shuts down gracefully, showing you live status and changelog updates as the bot restarts — no more silent hangs or lost context. We've also added voice transcription via Deepgram, so when you send audio messages, your agent automatically converts them to text for better understanding. On setup, your agent gains standard tools and full onboarding permissions from day one, and the first onboarding step now guides you through group setup rather than format choice — plus simple mode is on by default so you see concise reasoning summaries right away.

WAM Agents 0.15.0 adds the /assist command for quick help and bot menu access in your language, plus a streamlined /simple toggle so you can switch reasoning detail on or off without separate commands. You'll also see a handy onboarding checklist when you create a new agent, and TodoWrite todos now display in the /simple spinner mode so you stay informed even in condensed view.

WAM Agents 0.14.0 now gives you fine-grained control over how much detail you see during agent reasoning — use /simple to get just the essentials, or /detail to watch the full thinking process unfold. Behind the scenes, model-switch commands work reliably everywhere you chat, errors from tool calls no longer crash your session, and Markdown formatting that Telegram can't parse gets silently cleaned up so you always see readable text instead of raw symbols.

Your agent now captures detailed usage analytics during conversations, giving you clearer insights into how your sessions are running — you'll spot patterns in command use, reasoning depth, and media handling that help you work more effectively with your assistant.

Your agent now has access to more powerful media and data processing tools built in — ffmpeg for audio and video handling, imagemagick for image work, jq for structured data parsing, and wget for file retrieval. This means your agent can handle richer file formats and more complex requests without delays.

WAM Agents now logs tool failures from structured data streams, giving you better diagnostics when something goes wrong during agent reasoning. This means troubleshooting conversation issues becomes faster and clearer — you'll see exactly where processing stumbled instead of just a generic error.

WAM Agents now batches incoming media into albums, so when you send multiple photos or voice messages in quick succession, your agent receives them as a single grouped unit rather than separate items — making conversations with visual or audio content feel more natural and cohesive.

You can now switch your agent's reasoning model on the fly with /opus, /sonnet, /haiku, and /mythos commands — pick the depth and speed that fits each conversation. This release also syncs the version history and updates the release workflow so you get updates without delays.