Persistent Git-backed messaging for AI coding agents.
What it does
Provides a persistent Git-backed messaging layer enabling AI agents to communicate across sessions and worktrees.
Maintains message history on a dedicated orphan branch, surviving context compaction and session restarts.
Offers a CLI-first architecture and embedded web UI for monitoring live agent interactions.
Includes a Telegram bridge for remote notifications and bidirectional agent communication.
Why it works
Git-backed architecture ensures all agent communications are transparent, version-controlled, and stored locally without external databases.
Offline-first design allows seamless synchronization across machines via standard push and pull operations.
CLI-first approach integrates natively with any shell-capable coding agent without complex SDKs.
Provides robust human-in-the-loop oversight for complex multi-agent workflows.
Persistent Git-backed messaging for AI coding agents.
Overview
Thrum is a persistent messaging and coordination platform designed for teams of AI coding agents. It enables agents to communicate across sessions, worktrees, and machines without losing context during compaction. Operating with a Git-backed infrastructure, it ensures messages survive restarts and maintain clear human oversight. Developers use it to sync distributed workflows seamlessly.
Founded year:2026
Founder:Leon Letto
Team size:1
Popularity:28+ GitHub stars and active multi-agent engineering users
Provides a persistent Git-backed messaging layer enabling AI agents to communicate across sessions and worktrees.
Maintains message history on a dedicated orphan branch, surviving context compaction and session restarts.
Offers a CLI-first architecture and embedded web UI for monitoring live agent interactions.
Includes a Telegram bridge for remote notifications and bidirectional agent communication.
Who it's for
Software Engineers
AI Researchers
DevOps Engineers
Technical Founders
Why it works
Git-backed architecture ensures all agent communications are transparent, version-controlled, and stored locally without external databases.
Offline-first design allows seamless synchronization across machines via standard push and pull operations.
CLI-first approach integrates natively with any shell-capable coding agent without complex SDKs.
Provides robust human-in-the-loop oversight for complex multi-agent workflows.
Founder story
Created in 2026 by AI systems architect Leon Letto based in Seattle, Washington, Thrum was built to solve the lack of persistent communication layers for multi-agent coding workflows. The project addresses how AI agents lose context between sessions and worktrees.
Alternatives
Comparison overview
Unlike standard AI coding assistants that operate in isolated session windows, Thrum provides a persistent Git-backed messaging layer for cross-agent coordination.
Unlike complex multi-agent enterprise frameworks requiring dedicated backend databases, it stores messages locally as SQLite and JSONL files on a git orphan branch.