Local-first agentic trading, minimum tokens | Built by @AndreaPN | 404AI Labs
Joined April 2023
- Tweets961
- Following60
- Followers146K
- Likes12.3K
Best moment: after the fix i asked claude code for one more test. it resumed the same session and wrote the test that "would have caught the original bug",remembered the whole context.
Codex then did a read-only review. no files touched.
@bcherny @thsottiaux your agents did all the actual work here, curious what you'd want a host app like this to never do
AgentOS v2026.10.20 ⚙️
🖥️ Live acp.status and streaming by default for coding harness runs
🔁 Finished harness children stay idle, follow-ups resume the same session
🛑 Stops are final: parents are told not to restart a run you stopped
📄 Completions carry the final answer, long answers page in full
💸 Resume cost keyed per ACP session, no more double charging
⚡ Gateway no longer stalls requests behind next-prompt suggestions
📊 Exact 18-decimal base units in the PnL ledger, cleaner Trade tab rows
built for: AI agents
AgentOS v2026.10.19 🧢
🔄 Restart the gateway from chat with /restart
💬 Works from web, desktop, Telegram, Discord and Slack
🤖 The agent can restart the gateway via the gateway tool
⏱️ Graceful 30s turn drain, or force to cancel instantly
🧩 In-place re-exec keeps launchd, systemd and desktop tracking one process
🛠️ gateway restart --force and upgrade now restart unmanaged gateways
📋 Copy token contract addresses from the Desktop Trade tab
Hand coding work to a real coding agent 👨💻⚙️
Ask in chat: "have Claude Code fix the failing test in ~/code/app". AgentOS starts Claude Code, Codex, Gemini CLI or 20+ others over ACP as a child session.
By default, every file edit or command waits in your approval queue. When the harness is done, the result comes back to your chat with runtime, tokens and cost.
Persistent sessions, git worktrees, resume and cancel are built in.
First run: `agentos acp doctor claude --probe` to check your setup.
AgentOS v2026.10.16 🧢
🤖 Delegate coding to Claude Code, Codex, Gemini CLI, Cursor & 20+ agents via ACP
⚡ Parallel tool calls with live results and batched approvals
🪶 Deferred tool loading: 63 → 40 tool schemas per request
💾 Whole-conversation prompt caching with optional 1h TTL
🧠 Unified compaction that also runs ahead of time when idle
🎚️ Effort by router tier and per-turn timing traces
🛡️ Sturdier streams and reasoning replay across tool calls
AgentOS: a local-first agent, built fully in the open 💪
📦 39.8K PyPI downloads
🚀 70 releases in 92 days
🔧 2,652 commits · 1,016 merged PRs
👥 40 contributors · 53 forks
✅ 1,251 issues closed, 9 open
Still 66 ⭐. Help us change that.
Star it, fork it, ship a PR 🧢
👉 useagentos.dev
AgentOS v2026.10.13 🧢
💡 Suggested next prompt as ghost text: Tab to take, Enter to send
⏳ Short Retry-After waits stay on the same model, no needless failover
🧠 Adaptive thinking for Claude 4.6+ and Claude 5 models
🛑 Normalized stop reasons: refusals, max_tokens & pause_turn handled right
⚡ Tool calls and thinking stream instantly, no replayed responses
🔌 MCP: no tool-name collisions, real errors, untrusted output wrapped
🎨 Readable light-theme code blocks on Desktop & web console
"usage follows utility"
The ticker is $AGENTOS
"Usage follows utility." @useAgentOS is a local-first AI agent that natively supports agentic trading + tokenized RWA, and it's being built on Robinhood Chain. Ready for users to install and use today. 🧢
Start saving on inference costs today by setting up your own personal AI agent with AgentOS. It comes with Pilot Router support by default, so you can route requests smarter and keep spend under control.
I saved $124.86, almost 94.7%, by using AgentOS with Pilot Router. It felt like finding extra liquidity in my wallet with zero stress. How about you? 👀
Check by running: `agentos cost savings --pdf <file_path>`
Most people use AI agents like a genius with amnesia: fix 3 mistakes today, fix the same 3 tomorrow.
The fix: make every correction stick.
Correction → rule
Workflow → skill
Failure → eval
Here's how to set up a learning flywheel in AgentOS 👇
GM AgentOS holders, you can use the playbook below to learn how to arbitrage $AGENTOS and chAGENTOS and take profit step by step.
Volatility farming chAGENTOS - WETH on @ChopLogsApp
When $AGENTOS moves on the main pool, the chAGENTOS-WETH pool lags behind. Arbers close the gap with wrap/unwrap + swaps, and every arb generates fees.
More volatility → more arbs → more fees → higher farm APR paid by $CHOP token.
Wrap, LP, farm in 2 min 👇
choplogs.app/logs/0xa6308e0b…
chAGENTOS - WETH volatility farming APR is extremely high 🚨🚨🚨
🔥 Farmer APR: 202%
📉 chAGENTOS trading 6.4% below backing
(1 chAGENTOS = 1.0034 AGENTOS)
LPs → add liquidity & farm the volatility
Arbers → the gap is open on the @ChopLogsApp
Who's farming with us? 👇
AgentOS is now partnered with @ChopLogsApp to set up the chAGENTOS log on Chop 🔥
More liquidity and a fresh use case for $AGENTOS.
Takes about 2 minutes:
> Wrap your $AGENTOS into $chAGENTOS
> Add LP chAGENTOS-WETH
> Start farming, reward paid by $CHOP token
Let's go: choplogs.app/logs/0xa6308e0b…
P.S: we are building Chop skill for AgentOS 💪
AgentOS retweeted
Volatility farming chAGENTOS - WETH on @ChopLogsApp
When $AGENTOS moves on the main pool, the chAGENTOS-WETH pool lags behind. Arbers close the gap with wrap/unwrap + swaps, and every arb generates fees.
More volatility → more arbs → more fees → higher farm APR paid by $CHOP token.
Wrap, LP, farm in 2 min 👇
choplogs.app/logs/0xa6308e0b…
AgentOS is now partnered with @ChopLogsApp to set up the chAGENTOS log on Chop 🔥
More liquidity and a fresh use case for $AGENTOS.
Takes about 2 minutes:
> Wrap your $AGENTOS into $chAGENTOS
> Add LP chAGENTOS-WETH
> Start farming, reward paid by $CHOP token
Let's go: choplogs.app/logs/0xa6308e0b…
P.S: we are building Chop skill for AgentOS 💪