ShipitAi

Open-source Python agent runtime ⚡ 50+ tools, any LLM (Claude · GPT · Gemma), MCP, guardrails, computer use + a Claude-Code-style CLI → pip install shipit-agent

[ShipitAi @shipit_ai]

Shipped ShipIt Forge v2.2.0 — and the best part wasn't a feature. It was watching it review its own pull request and tell us we were wrong.

ShipIT Forge is an open-source autonomous coding agent for GitHub. It fixes issues, opens PRs, reviews code with a security lens, and auto-fixes CI. You bring your own LLM key — 9 providers supported.

v2.2.0 adds three deterministic security scanners that run on every pull request, before the model does: committed credentials, infrastructure (Dockerfile, Kubernetes, Terraform, workflows), and the source code itself — injection, path traversal, credentials written to logs, unsafe deserialization. No model call. No token cost. The same answer every time.

But the hard part of a scanner isn't finding things. It's not crying wolf.

A scanner people stop reading is worse than no scanner at all. So most of the work went into what it doesn't say:

- Taint is matched against the code on a line, not its prose. Our own log message ending "using the workflow token" contains the word "token" and leaks absolutely nothing. It was the first false positive the scanner ever produced — on its own author.

- Comment-only lines are skipped. The rule table documents the pattern it looks for, and promptly reported itself.

- Findings in test files drop to low severity instead of vanishing. A test suite has to contain the thing it detects. But a credential pasted into a test is still a credential, and quietly dropping it is how one stays there.

On our own repository that took the report from 15 findings — 8 of them its own test fixtures — down to zero blocking.

Then we opened the pull request, and Forge reviewed it. It flagged that we'd granted checks: write to every job in the workflow when only one job publishes a check run. It was right. We scoped the permissions per job and pushed the fix.

An agent that only ever agrees with you is a very expensive linter.

MIT licensed: https://t.co/KlW9toZkm0
https://t.co/yn46r13euZ

[ShipitAi @shipit_ai]

AI agents are easy to demo.

Reliable AI agents are hard to ship.

That’s why I created ShipIt Agent. 🚀

ShipIt is an open-source, production-grade Python runtime for building AI agents that can reason, use tools, maintain memory, verify their work, and operate across multiple LLM providers.

What makes ShipIt different?

✅ Structured output with in-conversation auto-retry ✅ Verifier networks for process supervision ✅ Principled episodic memory ✅ Time-travel replay for agent debugging ✅ Computer-use agents ✅ Multi-agent crews and deep-agent workflows
Built for developers who want flexibility, transparency, and control:
🧪 1,800+ passing tests 🤖 11 LLM providers 🛠️ 30+ built-in tools 🧠 40+ ready-to-use agents 🔓 Open source and MIT licensed
Install ShipIt:
pip install 'shipit-agent[all]'
📘 Documentation: https://t.co/vgLjc65M5m

⭐ GitHub: https://t.co/dhX9UHXgvv

If you’re building AI agents, I would genuinely appreciate your feedback. Try ShipIt, explore the repository, and let me know what you think we should build next.
#ShipIt #AIAgents #AgenticAI #OpenSource #Python #GenerativeAI #LLM #DeveloperTools #ArtificialIntelligence

[ShipitAi @shipit_ai]

Shipit agent 1.7 — "the working set" !!

A provider-agnostic Python agent runtime that reaches for a tool only when it needs one — and burns far fewer tokens doing it.

Deferred tool loading (core resident, the rest paged in on demand)

- Files straight in: images, PDFs, code
- MCP + powerful sub-agents
- Improved Live streaming event feed
Same code, every provider.

https://t.co/mI3ys7QO7Q

[ShipitAi @shipit_ai]

ShipIt Forge v2.2.0

- https://t.co/KlW9toZkm0
- https://t.co/yn46r13euZ

[ShipitAi @shipit_ai]

SHIPIT Agent 💻 !!

https://t.co/NL1m3Iebv6

[ShipitAi @shipit_ai]

@RoundtableSpace SHIPIT Agent, Powerful Python agent runtime with tools, MCP, Hooks, Skills, Rag, memory, sessions, reasoning, and streaming packets.

Give this project a ⭐ if you like it!

https://t.co/NL1m3Iebv6

[ShipitAi @shipit_ai]

@Nozelcode SHIPIT Agent, Powerful Python agent runtime with tools, MCP, Hooks, Skills, Rag, memory, sessions, reasoning, and streaming packets.

Give this project a ⭐ if you like it!

https://t.co/NL1m3Iebv6

[ShipitAi @shipit_ai]

@lightsilver323 SHIPIT Agent, Powerful Python agent runtime with tools, MCP, Hooks, Skills, Rag, memory, sessions, reasoning, and streaming packets.

Give this project a ⭐ if you like it!

https://t.co/NL1m3Iebv6

[ShipitAi @shipit_ai]

SHIPIT Agent, Powerful Python agent runtime with tools, MCP, Hooks, Skills, Rag, memory, sessions, reasoning, and streaming packets.

Give this project a ⭐ if you like it!

https://t.co/dhX9UHXgvv

[ShipitAi @shipit_ai]

⚡ shipit-agent v1.0.18 — CLI is here

$ shipit → full agent REPL in your terminal
🛠 shipit code — fixes your repo (approval cards, diffs, runs your tests)
🌐 shipit browse — drives a real browser
🔌 --mcp playwright · guardrails · self-healing tool calls

pip install -U shipit-agent
⭐ https://t.co/hcWAKiemsF

[ShipitAi @shipit_ai]

🚀 SHIPIT Agent v1.0.17 is live — an open-source Python runtime for tool-using AI agents.

⚡ Any LLM: Claude, GPT, Gemma 4, Bedrock
🛠 50+ tools, MCP catalog, guardrails, computer use
💻 Claude-Code-style CLI: shipit code / browse / serve

pip install shipit-agent
📖 https://t.co/nRlM4PMwSv