Adi

@opencrabs 🔥🦀🦀🔥

v0.3.70 JUST DROPPED 🦀🔥 🧠 Kimi reasoning-effort: per-model config, validated against allowed set 📏 Kimi plan-tier drives context window (API vs Coding) 🔒 Skills review_gate: frontmatter for high-stakes skills needing approval 📋 Plan discard: agent-initiated now requires autonomy, user always allowed 🎯 Plan yolo: design-track refusal scoped to /plan slash origin 🔧 MiniMax live model fetch from /v1/models endpoint 🪟 Mission-control Flakiest panel: windowed so fixed tools age out 👻 Phantom detector: language-agnostic, accents no longer disable it ⚡ Lazy-tools perf: LRU-bound active set, tool_search on use 18 COMMITS • 9 FIXES • 5 FEATURES • 5,017 tests Get it 🧵👇

@AdolfoUsier 🔥🔥🔥

OpenCrabs v0.3.69 is out. This one is about broadening the model lineup and making the chat feel more alive to interact with. The headline is a built-in Moonshot AI provider. Kimi was reachable before only as a custom provider, which meant fiddling with base URLs by hand. Now it's first-class. During onboarding you pick between the API plan on api.moonshot.ai and the Coding plan on api.kimi.com, which is the Kimi token subscription, and the wizard fetches the live model list for whichever plan you chose. K3 and the coding models show up automatically. The other big one is follow-up suggestions. When the agent finishes a turn it can offer a few ready-to-send next steps. You tap one and it fills your input box instead of firing immediately, so you can edit before sending. This works in the TUI and on all four chat channels: Telegram, Discord, Slack, and WhatsApp. It is a passive convenience, never a question you have to answer. MiniMax-M3 is now wired through the whole stack. The pricing row and the wizard already knew about it, but the config example, README, crate docs, and the usage normalizer all still stopped at M2.7. Now M3 is the default and the normalizer collapses it to one ledger row so /usage stays clean. There's also click-to-expand in the TUI. Contribution of @moneyacademyKE. Before, Ctrl+O flipped every collapsed block at once. Now you can left-click a single tool-call or reasoning block to expand just that one. Ctrl+O still does the all-at-once thing. Rounding it out: Kimi K3 now prices correctly on /usage instead of billing zero, a new Command Code CLI provider, contribution of @syafriadit1, and two display-name cleanups so the providers read as Moonshot AI and Anthropic. Thank you for Adi and Money Academy for the amazing contributions. 19 commits. 12 features. 4 fixes. 5,018 tests. Grab it at opencrabs.com

@opencrabs Goat... 🔥🔥🔥

🦀 Get v0.3.69 Single Rust binary recommended: 🐙 github.com/adolfousier/op… cargo install opencrabs 📦 crates.io/crates/opencra… Built-in Moonshot AI provider, tap-to-send follow-up suggestions across every channel, MiniMax-M3, and click-to-expand. Grab it at opencrabs.com

v0.3.69 JUST DROPPED 🦀🔥 🌙 Built-in Moonshot AI (Kimi) provider 🔀 API plan + Coding plan endpoint picker 💬 Tap-to-send follow-up suggestions on TUI, Telegram, Discord, Slack, WhatsApp 🧠 MiniMax-M3 wired in across the stack 🖱️ Click a tool or reasoning block to expand it 💰 Kimi K3 now prices correctly on /usage 🖥️ Command Code CLI provider 19 COMMITS • 4 FIXES • 12 FEATURES • 5,018 tests Get it 🧵👇

@AdolfoUsier @Kimi_Moonshot @opencrabs

Holy moly🥹🔥! Kimi did this animation with 1% of Allegretto plan usage. But hoo boy it did worth it! @Kimi_Moonshot cooked hard on this one. A perfect match to @opencrabs.

@opencrabs 🔥🔥🔥

🦀 Get v0.3.67 Existing users: Run /evolve to update. Single Rust binary recommended: 🐙 github.com/adolfousier/op… cargo install opencrabs 📦 crates.io/crates/opencra… Plan lifecycle engine, multilanguage analyzer, flow chrome with plan title and checklist progress, web search retry, Telegram rate-limit handling. Grab it at opencrabs.com

v0.3.67 JUST DROPPED 🦀🔥 Huge shout out to @leshchenko1979 and @syafriadit1 for the great contributions. 🧠 Full plan lifecycle engine with design and checklist tracks 📋 /plan, /show-plan, /execute, /discard on TUI and Telegram 🌍 Multilanguage prompt analyzer with 6 language packs 🔄 Web search rotates User-Agent and retries DuckDuckGo 403s 💬 Telegram retries on 429 rate limits 📊 Flow chrome shows plan title, checklist progress, and goals 🔧 Config hot-reload now surfaces recovery to the user 🎯 Channel context headers include chat_id and thread_id 🗑️ Dropped orphaned task_manager and SQLite plan path 52 COMMITS • 20 FIXES • 20 FEATURES • 4,938 tests Get it 🧵👇

OpenCrabs v0.3.66 is out. This release is a deep pass on two things: how you switch models, and how the Telegram flow block behaves while a turn is running. A lot of it came from watching the bot mishandle mid-turn messages and fixing each case properly. Model switching got a real overhaul. You can now type /models and switch directly on any channel, no menu walk. There's a scope selector so you choose whether the switch is just this session or global, and a headless session set-model command for scripting. On reload, force_default pushes the default provider/model pair to every session so nothing drifts silently, and startup now warns you when a config value has drifted instead of failing quietly. The flow block is the other half. Final responses on Telegram now render as native rich blocks, which killed the code-fence mangling that had been chewing up tables and snippets. The live header got a proper pipeline: wall-clock duration, finished/failed/timeout states, and bash line-start comments surfaced as live status. Then we reordered it so the status message leads, bold, with the tool count and duration trailing in italic. That last one took two tries because the header was built from two pieces in different places and the first attempt reordered the wrong pair. Mid-turn message handling got a cluster of fixes. If the bot asks you a question with buttons and you type an answer instead of tapping, that text now resolves the question directly instead of sitting there until timeout. A slash command that lands mid-turn injects as its own directive and actually runs, instead of getting folded into whatever the bot was already doing. And turns are now claimed atomically, so a follow-up or a reaction can't fork a second parallel turn. RSI learned to propose. It watches for recurring tool sequences and offers them as skill candidates, and repeated user requests become suggested slash commands. It also stops re-proposing skills you've already applied or rejected, which was getting noisy. Under the hood, the entire Telegram handler got modularized. The #471 work split one giant file into focused modules for delivery, resume, media, markdown, keyboards, and intermediate sends, with a shared final-response path so the classic and rich renderers can't drift apart. Plus a per-group open mode for trusted groups, retryable classification for transient model errors, and a clean pass on the new Rust 1.97 clippy lints. 53 commits. 13 features. 23 fixes. Grab it at opencrabs.com

v0.3.66 JUST DROPPED 🦀🔥 🔀 Direct /models provider/model switching on every channel 🎨 Native rich blocks on Telegram, fence mangling gone ⏱️ Flow header leads with the status message, then count and duration 🔓 Opt-in per-group open mode for trusted groups 🧠 RSI proposes skills and slash commands from your own patterns 💬 Type a text answer to a pending button question 🎯 Slash commands landing mid-turn actually run now 🔒 Atomic turn claims, no forked parallel turns ♻️ Telegram handler split into focused modules 53 COMMITS • 23 FIXES • 13 FEATURES Get it 🧵👇

v0.3.63 JUST DROPPED 🦀🔥 💬 Discord gets full feature parity with Slack 📎 doc_gen embeds images inline in PDF and DOCX 📋 Telegram flow logs via rich API (32K chars) 📊 /usage breakdown by provider and model 👁️ image.vision.provider override for vision-only providers ⚡ Concurrent auto-approved tool batches 🎨 PPTX/DOCX/XLSX style layers with brand templates 🔒 Security deny-by-default access model Should out to our great contributions once again @leshchenko1979 and @syafriadit1, thank you! 35 COMMITS • 13 FIXES • 13 FEATURES • 4,680 tests Get it 🧵👇

v0.3.61 JUST DROPPED 🦀🔥 📋 Grouped tool calls collapse into one expandable block 🔽 Intermediate steps fold into an in-place processing log 👏 React to Crab and it reacts back, no full turn spun up ⚡ Mid-turn reactions steer the running loop live 🎯 Self-goaling: Crab sets and drives its own multi-turn goals 🔔 Rebuild outcomes reach whoever asked, on any surface 📊 Excel/spreadsheet parsing (XLSX, XLS, CSV) 🧹 Cleaner Telegram rendering, formatted even when expanded 27 COMMITS • 16 FIXES • 6 FEATURES • 4,563 tests Get it 🧵👇

🦀 Get v0.3.60 Single Rust binary recommended: 🐙 github.com/adolfousier/op… cargo install opencrabs 📦 crates.io/crates/opencra… Native web_scrape tool with zero AI cost, 4x faster than Claude CLI, SSRF protection, profile-aware export. 8 features, 4 fixes. Grab it at opencrabs.com

v0.3.60 JUST DROPPED 🦀🔥 🌐 Native web scraping: URL-to-markdown with zero AI cost, 4x faster than Claude CLI 🔒 SSRF protection: blocks private IPs, localhost, cloud metadata, non-http schemes 🧹 Structural HTML cleaner: strips scripts, styles, inline handlers, collapses whitespace 🎯 Main-content extraction: CSS selector cascade finds the article, not the chrome 📝 HTML-to-markdown: htmd converts clean HTML, absolutizes URLs for portable images ⚡ JS-shell detection: reqwest fetch with browser escalation for React/Vue/SPA pages 🗺️ Sitemap discovery: recursive URL enumeration with 1000-URL cap and depth limits 📁 Profile-aware export: markdown files land in project directory, not temp 🎬 Vision routing: video/GIF analysis works with any vision provider, not just Gemini 👋 Personalized greeting: /evolve wake-up message adapts to your context 17 COMMITS • 4 FIXES • 8 FEATURES • 4,507 tests

@AdolfoUsier @opencrabs oke sir, i dm

Bro I am super sorry about that. I really want to help you fix that ASAP. Just to confirm. 1. do you have debug logs enabled with --debug flag? If yes you can or ask crabs to debug logs and make a report so you can share with me or share the logs with me by DM so I can debug it. I can also create a group on telegram with you and your bot and opencrabs so we can tackle this together. 2. you are on version .58, right? 3. did you follow this exact steps? #troubleshooting" target="_blank" rel="nofollow noopener">docs.opencrabs.com/channels/whats…

v0.3.55 JUST DROPPED 🦀🔥 📁 Native /profiles dialog: browse, switch, create, delete, migrate from TUI 💬 /profiles now works in Discord, WhatsApp, Slack 🔧 Tools.toml auto-recovers from .bak snapshot on parse failure 🔑 Xiaomi endpoint_type selector: API vs Token Plan per provider 🧠 Telegram replies fetch exact message by platform ID 👤 Reply context includes the author's name now 🧪 49 new tests for profiles dialog flows 📊 README and TESTING.md brought up to date 20 COMMITS • 5 FIXES • 10 FEATURES • 4,257 tests

@AdolfoUsier @opencrabs It seems like the issue is with my VPS, so try debugging OpenCrabs yourself.

What do you mean worked with different method? Sessions do persist and you can confirm that hitting /sessions on TUI, you will see the whatsapp sessions there. If its restarting WhatsApp connection on restart that means the connection setup was not complete so it didnt persist. Are you running daemon or TUi right now? Also if you are using OpenCode CLI it have limitations and cant use all opencrabs tools.

@AdolfoUsier @opencrabs Okay, it worked, even though I used a different method. And here’s a little something I found in the open code, though I’m not sure if it’s actually correct.

@syafriadit1 @opencrabs You are welcome brother. Please confirm if is it all working fine now? Thank you

@AdolfoUsier @opencrabs oke tks sir

If you want to keep TUI always on instead of Daemon(background process). # 1. Install tmux sudo apt install tmux # Debian/Ubuntu # or sudo yum install tmux # CentOS/RHEL # 2. Find where opencrabs is installed which opencrabs # or if not in PATH: find / -name "opencrabs" -type f 2>/dev/null # 3. Start a named tmux session tmux new -s opencrabs # 4. Run opencrabs inside it opencrabs # (or full path from step 2: /usr/local/bin/opencrabs) # 5. Detach (leave it running, close SSH safely) # Press: Ctrl+B, then D # 6. Reconnect later (even after closing SSH) tmux attach -t opencrabs That's it. TUI stays alive the whole time, even if you disconnect. Reconnect anytime with tmux attach. 🦀

@AdolfoUsier @opencrabs Now I'm facing another classic issue: responses on the Telegram and WhatsApp channels stop if I close the OpenCrabs terminal on the VPS. 😅

@syafriadit1 @opencrabs Cool thanks Adi, thank you! Is it working after that? Cheers brother

@AdolfoUsier @opencrabs Oke done sir

Got you, please reset, clear all devices, connect just one. Let me know how it goes, please follow instructions as both WA devices and /onboard:channels needs to be reseted to clean state. - 1: First remove all connected devices on your whatsapp app, including and mandatory to remove also ```opencrabs``` device you just added on whatsapp app device list - 2: then go back to /onboard:provider and hit ```R``` to reset connection - 3: once reset is confirmed, go to whatsapp, settings,. devices and link a device from scratch. --- Sidenote: 1. You talk to your self very much like Telegram DMs. 2. If anyone message that paired number and is on allowlist, it should reply to them. CRITICAL: Always reset connection on /onboard:channels with "R", and remove devices on whatsapp/settings, before connecting new numbers, it will only take last anyways.

@AdolfoUsier @opencrabs Okay, but number 2 is the one that isn't in effect yet.

But super happy to see it working for you on that number, keep in mind WhatsApp is not like Telegram, we don't have a bot, so its your own number. 1. You talk to your self very much like Telegram DMs. 2. If anyone message that paired number and is on allowlist, it should reply to them. CRITICAL: Always reset connection on /onboard:channels with "R", and remove devices on whatsapp/settings, before connecting new numbers, it will only take last anyways. @syafriadit1

@AdolfoUsier @opencrabs But, no reply to them

@syafriadit1 @opencrabs Its not a bot like Telegram but its connected to your own number, so that person message yourself, it should reply to them.

@AdolfoUsier @opencrabs Sorry, Sir, I didn't realize it had already been activated. But strangely enough, I now have two active numbers. Is this normal, or is it an anomaly?

@syafriadit1 @opencrabs

@AdolfoUsier @opencrabs Lol. I tried adding another WhatsApp number to the allowlist to message this paired bot, but there was no response.

@AdolfoUsier @opencrabs This is a bit strange; the initial error occurred using WhatsApp Business on Android, so I tried pairing a different number using the standard WhatsApp app. However, it's still odd because the bot only responds to private messages from the specific number that was paired with it