Back to Ecosystem Pulse
ToolExperimental
thrum
by leonletto
Git-backed persistent messaging for coding AI agents
Go
Updated Jul 4, 2026
Share:
What It Does
Provides persistent, Git-backed messaging for coding-focused AI agents so conversations and state survive across sessions, worktrees, and machines. Uses a lightweight CLI and local storage (SQLite + git) to record agent interactions, message history, and artifacts, making coordination reproducible and auditable. Designed for multi-model setups (Claude Code, Codex, etc.), it ties chat logs to Git commits and worktrees for easy branching and review. Git-backed messaging
The Value Proposition
As agents collaborate on code, being able to reconstruct who said what and which artifacts resulted is critical for trust and debugging. Thrum makes agent interactions reproducible and traceable by versioning messages alongside code, which helps establish an agent track record Agent Registry Pattern. For agent-to-agent evaluation and reputation work, durable, source-controlled logs are a practical foundation for building trust signals and continuous evaluation workflows Evaluation-Driven Development (EDDOps).
When to Use
Developers coordinating multiple coding agents who need durable, reviewable message history tied to code and branches. Orchestrator-Worker Pattern
Applications
- Record and replay multi-agent coding sessions tied to specific git branches or worktrees
- Audit agent decisions and generated artifacts during code reviews or post-incident analysis
- Branch experimental agent workflows and compare outcomes by keeping chat history versioned with code
Works With
openaiclaudegitsqlitetmux
Topics
agent-coordinationai-agentsclaude-codeclicodexgitgolangllm-toolsmessagingmulti-agent+2 more
Similar Tools
autogenswarm
Keywords
multi-agent orchestrationagent interaction loggingagent track recordgit-backed messaging