guozijn

guozijn / agentd

Public

State Machine Daemon designed specifically to orchestrate and persist the state of AI agents

agentd.cc
18
0
89% credibility
Found May 24, 2026 at 21 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Rust
AI Summary

agentd is a background service that helps multiple AI assistants work together on complex tasks without getting in each other's way. It keeps track of multi-step projects, makes sure only one agent works on something at a time, handles situations where an agent gets stuck, and prevents agents from editing the same files simultaneously. The service stores everything in a local database so nothing is lost if something crashes, and it includes examples showing how AI agents can use it to coordinate their work.

How It Works

1
💡 You need AI agents to work together

You've got multiple AI assistants helping with a project, but they keep stepping on each other's toes.

2
🔧 You start a coordination service

You launch a small background program that acts like a traffic controller for your AI agents.

3
📋 You define a multi-step task

You create a task with several steps, where some steps must wait for others to finish first.

4
Your AI agents connect and start working
✅
Agent finishes step successfully

The agent completes its work and the service marks that step as done.

⏰
Agent takes too long or stalls

The service notices the agent stopped responding and makes that step available for another agent to try.

5
🔒 Agents coordinate file access

When an agent wants to edit a file, it asks the service for permission, preventing conflicts.

6
📊 You can check progress anytime

You ask the service how things are going and it tells you exactly which steps are done, running, or waiting.

🎉 Your coordinated agents succeed

All steps complete in the right order, with results saved and no conflicts between your AI helpers.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 21 to 18 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 agentd?

agentd is a state machine daemon written in Rust that acts as a durable coordination layer for AI agents. It runs out of process, persists task state in SQLite, and exposes a JSON Lines API over Unix Domain Sockets so your agent runtimes can coordinate long-running DAG work without becoming the source of truth themselves.

The daemon manages strict node states (PENDING, RUNNING, COMPLETED, FAILED), handles lease acquisition with heartbeat and timeout rollback, and provides resource locks for shared resources like repository files. Multiple AI providers can collaborate on the same codebase without stepping on each other.

Why is it gaining traction?

The hook is lease-based coordination with automatic timeout rollback. If a worker stalls, the daemon rolls the node back to PENDING and lets another worker pick it up. No manual cleanup, no orphaned processes.

Resource locks solve the multi-provider problem that keeps teams from running Claude, Codex, and DeepSeek agents together. You get exclusive file-level locks with metadata, conflict resolution without last-writer-wins, and automatic recovery when locks expire. The conflict resolution docs show real scenarios: two providers targeting the same file, how the coordinator decides which one proceeds.

The DeepSeek multi-agent loop example demonstrates a four-node DAG where agents generate and review worker scripts concurrently, then synthesize the results. That kind of pipeline orchestration is what you'd otherwise build from scratch.

Who should use this?

DevOps and MLOps teams running agent infrastructure who need durable coordination without Redis or a cloud service. Researchers building multi-agent pipelines where tasks span hours and must survive crashes. Teams using multiple AI providers on the same repository and needing provider-neutral conflict resolution. Individual developers tired of stateless agent scripts that lose everything on restart.

Verdict

At 18 stars and version 0.1.2, this is early-stage but the credibility score of 0.8999999761581421% reflects a well-structured codebase with comprehensive tests and schema migrations. The README is thorough, the IPC protocol is clean, and the lease/heartbeat mechanism handles the hard parts of distributed coordination.

If you need durable AI agent orchestration without external dependencies, this is worth evaluating. Watch for maturity in the ecosystem before betting production workloads on it, but the foundation is solid.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.