paradigmxyz

Multiplayer, self-hosted, secure agents.

98
14
89% credibility
Found May 21, 2026 at 99 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Centaur is a self-hosted AI agent platform that lets teams share one AI assistant instead of each person maintaining their own setup. Team members talk to the agent directly through Slack, and it works in isolated sandboxes with access to your code, tools, and internal services. The platform handles security by keeping credentials in a firewall layer that swaps them in-flight, so agents can use approved services without ever seeing raw passwords. It stores conversations and results durably so nothing is lost if you reconnect, and supports both quick one-shot questions and complex multi-step workflows that can pause and resume.

How It Works

1
đź’¬ Someone mentions the agent in Slack

A team member types a question or request directly to the Centaur bot in their Slack channel, just like messaging a coworker.

2
🤖 The agent gets to work in its own workspace

Centaur spins up a private workspace for this conversation where it can read files, run tests, and use approved tools without affecting anyone else.

3
đź”§ Your team's shared tools come alive

The agent uses Python tools your team has built once—like internal databases, APIs, or deployment systems—without ever seeing the actual passwords.

4
📊 Progress appears in the Slack thread

As the agent works, updates and results flow back into the Slack thread so the whole team can follow along in real time.

5
Different types of work, handled differently
⚡
Quick questions

One-shot answers that come back in seconds with the information you needed.

🔄
Complex investigations

Multi-step workflows that can run for minutes or hours, sleeping and resuming as needed.

âś… Your team has a shared AI coworker

Instead of everyone setting up their own AI tools, your whole team shares one agent that remembers context, follows your rules, and has access to your internal systems—all safely.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 99 to 98 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is centaur?

Centaur is a self-hosted platform for running shared AI agents across teams. You talk to an agent through Slack, and it spins up an isolated sandbox to actually do the work: run commands, inspect code, call tools, and ship results back to the thread. The platform handles sandbox lifecycle, durable state, credential injection, and workflow execution. It is built in Python with a FastAPI control plane, TypeScript for Slack integration, and runs on Kubernetes.

Why is it gaining traction?

The killer feature is the security model. Agents never see raw API keys—credentials are swapped in-flight by a firewall proxy. Each conversation runs in an isolated container with a default-deny network policy. You also get to run whatever agent harness you want (Claude Code, Codex, Amp) without being locked into one provider's ecosystem. The tool plugin system means your internal APIs become agent-accessible with a few lines of Python. Workflows let you build multi-step processes that survive restarts and can coordinate multiple agent turns.

Who should use this?

Platform teams building internal developer tooling will get the most value. If your org wants one shared agent instead of everyone running one-off local setups, this is purpose-built for that. Teams with existing Slack workflows who want to add AI-powered automation without sending data to third-party services will find the self-hosted model appealing. Organizations with strict security requirements around credential handling will appreciate the firewall-based secret injection.

Verdict

Centaur is a well-architected system with thoughtful security design, but at 98 stars it is early-stage and the credibility score of 0.9% reflects that. The documentation is solid for a project this size, but production readiness requires Kubernetes expertise and a 1Password-backed secret workflow. If you need self-hosted AI agents with strong isolation guarantees, it is worth a closer look—just budget time for the setup and treat it as a foundation to build on rather than a plug-and-play product.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.