SYNAPSE CHANNEL: The Local-First Coordination Bus for AI Agent Fleets

Overview

SYNAPSE CHANNEL is a local-first coordination substrate designed for fleets of independently running AI agents and humans working alongside them. It operates as a single WebSocket hub running entirely on the user's machine, serving as the authoritative source of truth for presence, file-scope work claims, chat, task status, agent capabilities, and resource offers. Unlike agent frameworks that orchestrate agents within a single process, SYNAPSE CHANNEL sits between them. It coordinates agents across processes and repositories without requiring a database, consensus protocol, or cloud service. The core installation is transport-light, relying on a single dependency and the Python standard library.

Benefits

SYNAPSE CHANNEL addresses common failure modes when multiple autonomous agents operate against shared work. These include collisions where agents edit the same files simultaneously, duplicated effort where agents pick up the same tasks, and lost handoffs where agents fail to signal the next step. The system provides several key advantages:

• File-Scope Claims:An agent leases a specific task and declares the files it will touch. The hub refuses overlapping live claims, ensuring two agents never edit the same file simultaneously.
• Shared Blackboard:A dual-ledger system tracks tasks, progress, and dependencies, allowing agents to see what is taken and what their work unblocks.
• Atomic Handoffs:Authority is transferred atomically between agents, eliminating windows where files are unowned.
• Durable Persistence:An append-only SQLite WAL event log replays on restart, ensuring live leases and task state resume rather than being lost.
• Local Operation:The hub runs entirely on the user's machine, avoiding the need for cloud services or complex consensus protocols.
• Protocol Agnosticism:The system connects existing agent protocols like MCP and A2A at the edge, allowing tools like Aider or custom scripts to participate in the bus without new code.

Use Cases

SYNAPSE CHANNEL is designed for teams that want to scale the number of agents they run without scaling the effort required to babysit them. It is particularly useful for:

• Multi-Agent Coding Teams:It allows fleets of coding agents to work in parallel within a single repository or across an ecosystem, ensuring that work claims are respected and handoffs are atomic.
• Human-Agent Collaboration:It supports humans working alongside AI agents, providing a shared view of presence, task status, and resource offers.
• Tool Integration:Existing tools like Claude Code, Codex, Cursor, and Aider can connect via the bundled server, exposing coordination verbs as tools.
• Local Fleet Management:The system provides a fleet cockpit, a read-only dashboard that offers a single view of the roster, claim status, progress, and task dependency graphs.
• Offline Operation:A deterministic rule-based fallback ensures operation even when offline or when the primary model worker is unavailable.

Pricing

The project is dual-licensed under AGPL-3.0-or-later, which is free for research, personal, and open-source use. A commercial license is available for closed-source and SaaS use. Specific pricing details for the commercial license are not provided in the available information.

Vibes

The project is currently in the 0.x pre-1.0 development release phase. The first stable commercial release, version 1.0.0, is planned with stable operational contracts and support surfaces. The community and developers are focused on building a robust foundation for multi-agent systems, with a roadmap that includes evolving the dashboard into a Studio operator control plane and planning a separate multi-machine Fleet layer to coordinate hubs across machines.

Additional Information

SYNAPSE CHANNEL ships as a single installable Python package. Installation is straightforward using pip. Optional extras include sqlcipher for page-level encryption of the durable coordination log, encryption for whole-file AES-GCM envelope helpers, and mcp for Model Context Protocol integration. The system uses an append-only SQLite event log in WAL mode for crash durability. Claims, releases, task updates, and checkpoints are recorded synchronously. On restart, the hub replays the log to rebuild state, ensuring live leases and task status resume without loss. For at-rest encryption of the live hub event store, the optional sqlcipher extra can be installed. This provides page-level confidentiality for the durable coordination log while the hub holds it open. Keys are managed via synapse encrypt-key commands, and operators must provide the key file to access the store. Governance surfaces like policy checks, approvals, ACLs, and Merkle roots are advisory by default but can be enforced. The hub records evidence and epistemic status for every action.