synapse-channel 0.45.0

Local-first coordination bus for AI agents across one repository or a whole ecosystem: work claims, presence, chat, a shared plan, and resource sharing over a WebSocket hub

Stop parallel AI coding agents from clobbering each other's files.
Local-first coordination bus — file-scope claims, a shared plan, and durable leases — for one repository or a whole ecosystem of them.

A local-first coordination bus for a fleet of AI agents working in parallel — within a single repository or spread across a whole ecosystem of them. One WebSocket hub is the shared source of truth for presence, work claims, chat, task status, and resource offers: agents address each other across projects and share one plan, while file-scope claims keep the agents in any one repository off each other's files.

The bus is transport-light (one dependency, websockets), hub-centric by design (one place owns presence, leases, and history), and runs entirely on the local machine. Model workers reply on-channel through any OpenAI-compatible endpoint, including a local Ollama server, with a deterministic rule-based fallback for offline use.

At a glance

graph LR
    A1["Agent"] --> H
    A2["Agent"] --> H
    A3["Worker"] --> H
    SUP["Supervisor"] --> H
    H["SynapseHub<br/>single source of truth"] --> CL["Claims & leases<br/>scope · epoch · checkpoint"]
    H --> BB["Blackboard<br/>plan + progress"]
    H --> CAP["Capabilities<br/>cards + routing"]
    H --> LOG["Event log (SQLite WAL)<br/>durable · replayed on restart"]

A claim leases a unit of work with a file scope, so two agents never edit the same files; the plan, handoffs, checkpoints, and a stall supervisor keep the work moving; and the durable event log means a hub restart resumes live leases rather than losing them.

Install

python -m pip install synapse-channel       # the release from PyPI
python -m pip install -e ".[dev]"           # or an editable dev checkout

This installs the synapse command. To run the hub as an always-on local service or a container, see the deployment guide (a systemd user unit and docker compose are both included).

First 60 seconds

On a clean Python environment, verify the installed CLI before wiring agents into a real repository:

python -m pip install synapse-channel
synapse doctor
synapse demo
synapse quickstart-coding

synapse doctor reports local setup issues such as identity, hub exposure, and missing waiters. A brand-new machine may warn that no hub or waiter is running; that is expected before service setup. synapse demo starts its own local hub, drives a planner/worker coordination flow, and succeeds when it prints:

success: coordination demo completed

synapse quickstart-coding creates a temporary coding-fleet workspace, runs the same no-collision coding demo used by generated workspaces, removes the temporary workspace after success, and prints:

success: coding fleet demo completed

Releases

This package is developed in the open and dogfooded daily: a fleet of coding agents runs its own coordination on it, so problems surface in real use and are fixed quickly. Releases are therefore frequent and mostly small — fixes and hardening rather than churn. The wire protocol and the public Python API stay backwards-compatible within a major version; any breaking change is called out in the changelog.

If you need a fixed target, pin a version (synapse-channel==X.Y.Z); to get the latest fixes, track the newest release. Both are supported.

Quick start

Launch a hub plus one or two local model workers in one command:

synapse team

Then, from another terminal, watch the channel or send a message:

synapse listen --name USER
synapse send --name USER --target FAST "what is the status of TASK-1?"

Running pieces individually

synapse hub --port 8876
synapse hub --port 8876 --db ./synapse.db            # crash-safe: resumes leases + history on restart
synapse hub --port 8876 --relay-log ./feed.ndjson    # mirror the channel to a compact file for observers
synapse worker --name FAST --provider ollama --model gemma3:4b
synapse worker --name OFFLINE --provider rule        # no network, canned replies
synapse worker --name TIER --provider tiered --model small --heavy-model big  # route trivial→rule, hard→heavy
synapse relay ./feed.ndjson                          # decode and print that file as readable lines
synapse ingest ./synapse.db --memory --cursor ./mem.cursor  # stream durable memory events since a seq cursor (NDJSON)
synapse board                                        # print the shared task/progress blackboard
synapse task declare BUILD --title "compile"         # declare/update the shared plan from the CLI
synapse task update BUILD --status done              # mark a plan task done so dependents unblock
synapse supervisor --idle-seconds 300                # LLM-free: re-offer tasks that stall
synapse manifest                                     # print the capability cards agents advertised
synapse a2a-card --endpoint-url https://agent.example.com/a2a/v1  # emit A2A Agent Card JSON
synapse a2a-serve --endpoint-url http://127.0.0.1:8877             # run the HTTP+JSON A2A bridge
synapse doctor                                       # check for common misconfigs (identity, exposure, hub, waiter)
synapse demo                                         # installed self-check: local hub + planner/worker flow
synapse quickstart-coding                            # create a temporary coding fleet workspace and run it
synapse new coding-fleet ./demo-fleet                # scaffold a runnable two-agent coding demo workspace
synapse hub --host 0.0.0.0 --token s3cret            # require a shared secret when binding off-loopback
synapse hub --max-connections-per-host 4             # cap simultaneous sockets from one remote host
synapse send --token s3cret --name USER "hello"      # agents present the token to a secured hub

Use it with your coding agent

Synapse coordinates the agents you already run; it does not replace them.

• Claude Code / Claude Desktop / Cursor (MCP): point the host at the MCP server and every coordination verb shows up as a tool — no Synapse-specific code.

  pip install 'synapse-channel[mcp]'
  synapse mcp --uri ws://localhost:8876        # add this to the host's MCP server config
  

• Aider, or any non-MCP tool: claim a file scope before editing and let a git hook release it on commit, so two sessions never touch the same files.

  synapse quickstart-coding                    # optional: run a temporary no-collision coding demo
  synapse new coding-fleet ./demo-fleet        # optional: keep the generated workspace
  synapse git-init --name aider-1              # one step: install the hooks + write the conventions guide
  synapse git-claim AUTH --paths src/auth --name aider-1
  aider src/auth/*.py                          # ... edit; the post-commit hook releases the claim
  

• Check the wiring: synapse doctor reports the common setup mistakes — no live waiter, a hub exposed without a token, an accidental identity — each with its fix.

• Install the always-on local services: synapse init prints or installs the hub, project presence, and non-LLM wake listener units. doctor --fix prints the exact commands when a waiter is missing.

  synapse init --project myrepo --identity myrepo/worker --install-user-services
  synapse init --project myrepo --identity myrepo/worker --start-user-services
  synapse doctor --fix
  

• Launch a provider command with a wake sidecar: worker-session exports the identity variables, starts a cheap syn arm listener, runs your command, and stops the listener when the command exits.

  synapse worker-session --identity myrepo/worker -- codex --sandbox danger-full-access
  

Agent ergonomics — the syn commands

For the short loop an agent runs every session — arm a waiter, send a message, read the inbox, glance at the board — the package also ships syn, a thin, identity-correct front end over the commands above:

syn name                          # resolve and print this terminal's identity
syn arm                           # keep a directed-only waiter armed (named <project>-rx, distinct from the sender)
syn say REMANENTIA,CEO "ack"      # send to one, several, or all
syn inbox                         # print messages addressed to you since the cursor
syn board                         # the shared task/progress board
syn who --me                      # show whether this identity and its -rx waiter are online

The one thing it gets right that a hand-rolled shell alias does not is identity. The project is resolved from --project, then $SYN_PROJECT (or $SYN_IDENTITY for a project/<type>-<id> multi-agent identity), and the working directory only as a last resort — so a command run from the wrong directory does not silently coordinate as the wrong project, and an identity that looks accidental (the home directory, a system path) is flagged rather than used in silence. Set $SYN_PROJECT once per terminal and the identity is stable across tool calls. The syn who --me shortcut dispatches to synapse who --me --name <resolved identity>; it reports the identity's presence separately from its -rx waiter because presence is not a wake loop.

syn-name/syn-wait/syn-say/syn-inbox/syn-board aliases are installed too; syn-wait uses the same persistent auto-rearming path as syn arm.

To make fresh terminals connect automatically, install the shell hook once:

synapse install-shell-hook --shell auto

New Bash/Fish/Zsh terminals then export SYN_PROJECT/SYN_IDENTITY and keep a cheap synapse arm sidecar running. The hook does not silently join whatever git checkout the terminal happens to start in. It joins the neutral SYNAPSE_DEFAULT_PROJECT lane, or user when unset, unless you explicitly set SYN_PROJECT/SYN_IDENTITY or opt a repository in with .synapse/project:

mkdir -p .synapse
printf '%s\n' myrepo > .synapse/project

For legacy CWD-derived behavior, set SYNAPSE_AUTO_PROJECT_FROM_CWD=1 in that terminal. The hook also wraps common provider commands (codex, claude, gemini, agent, ask, ollama) through synapse worker-session, so cloud and local LLM sessions inherit the same Synapse identity without polling or manual arming. Set SYNAPSE_AUTO_CONNECT=0 to disable it for a terminal.

Durability

Passing --db backs the hub with an append-only SQLite event log (standard library, WAL mode). Every claim, release, task update, resource offer, and chat message is recorded, and the hub rebuilds its state by replaying the log on start-up. The guarantee is split honestly by workload: the lease/claim path commits at synchronous=FULL (durable across an OS crash); the high-volume chat/history path commits at synchronous=NORMAL (durable across an application crash, may lose the last commit on power loss).

Token-thrifty observation

--relay-log mirrors every broadcast to a newline-delimited file in a compact short-key form (encode_lite), so a token-budgeted agent can watch the channel by tailing a file instead of holding a socket. synapse relay <file> decodes it back to readable lines and can resume from a saved --cursor. The lite form keeps the seven core envelope fields and drops auxiliary ones; the file is bounded by --relay-max-lines. A committed benchmark measures the saving honestly — see benchmarks/.

Exposure

By default the hub binds to loopback and runs with no authentication — the right posture for one operator on one machine. When that is not enough (a worker with tool-use, or a hub bound off-loopback), --token requires a shared secret that connecting agents present with --token. Binding off loopback without a token is refused rather than silently exposed: the hub will not start unless you set a token (and --metrics-token when metrics are on), or explicitly pass --insecure-off-loopback to accept the risk. This is a proportionate gate, not a cryptographic identity system.

MCP server face

Any MCP-compatible agent — Claude Desktop, Claude Code, an editor assistant — coordinates through Synapse with no Synapse-specific code. Install the optional extra and point the host at the command:

pip install 'synapse-channel[mcp]'
synapse mcp --uri ws://localhost:8876

synapse mcp runs a Model Context Protocol server over stdio that is itself a hub client, exposing the coordination verbs as MCP tools (claim, release, send, hand off, declare and update tasks) and the board, state, and manifest as live resources. The hub stays MCP-agnostic and the core install keeps its single dependency — see the MCP guide.

For Agent2Agent discovery, synapse a2a-card --endpoint-url ... projects the live capability manifest into an A2A Agent Card JSON document suitable for a thin HTTP edge to serve as /.well-known/agent-card.json. To run that edge directly, use synapse a2a-serve --endpoint-url ...; it serves the public Agent Card, forwards POST /message:send text/data/file parts into SYNAPSE chat, supports immediate POST /message:stream Server-Sent Events, exposes bridge-local task list/get/cancel plus push-notification configuration routes, accepts JSON-RPC 2.0 calls on /rpc, and can enforce Bearer auth plus request size/depth bounds, persist task state with --state-file, fail stale open tasks with --task-timeout, and bound one subscription wait with --subscribe-timeout. The bridge is intentionally a local-first HTTP+JSON edge: it stores bridge task state locally in owner-only state/temp files, rejects unsafe caller ids and webhook targets including delivery-time DNS or redirect targets that resolve to local networks, bounds stored tasks/history/artifacts/push configs/replay history with terminal-task retention GC, emits subscription replay only from the current bridge process, and does not claim independent A2A conformance until remote CI, interoperability, and real webhook receiver validation have run.

Git-native claims

A claim can be scoped to the git branch it happens on, resolved client-side:

synapse git-init                                 # one-step setup: install the hooks + write a .synapse/ guide
synapse git-claim TASK-1 --paths src/auth.py     # claim, tagged with your branch
synapse git-hook install                         # (git-init already does this) auto-release on commit/merge
synapse conflicts --check-diff                   # predict cross-branch merge conflicts

synapse git-init bundles the hook install with a short .synapse/git-claims.md onboarding guide (branch convention + worktree workflow). synapse state shows each claim's branch; installed git hooks release a claim when its files are committed or merged; and synapse conflicts flags two agents about to edit the same files on branches that merge into the same base. The hub stays git-agnostic — it stores the branch as opaque metadata and never runs git or reads a filesystem — so all git work is on the client. See the git-native claims guide.

Coordination model

1. Claim before you work: an agent leases a task by id; a live lease blocks other agents from claiming the same task.
2. Declare a file scope on the claim (a worktree and paths); the hub refuses a claim whose files overlap another agent's live claim — this is how two agents are kept off the same files. Agents in different worktrees never contend.
3. Leases auto-expire, so a crashed agent never holds a claim forever, and each lease carries an epoch so a superseded agent cannot act on a dead claim. An owner can save a durable checkpoint on the task; if its lease lapses, the next agent to claim the task inherits that checkpoint and resumes rather than restarting.
4. Release on completion; status and an optional artefact reference can be attached while the task is in progress. A held task can also be handed off atomically to another online agent — keeping its scope, status, and context, with no window for a third agent to grab it mid-transfer.
5. Presence, who, full state snapshots, and chat history are queryable at any time. After a reconnect to the same running hub, an agent can resume by idem_key (retried claims are not applied twice while the hub retains its idempotency cache) and a resume cursor (fetch exactly the messages it missed).

Alongside the lease registry, a shared blackboard holds the team's plan: a task ledger of declared work with dependencies (the hub refuses dependency cycles, so ready tasks are well-defined) and an append-only progress ledger a supervisor can read to spot stalls. A declared LedgerTask is the plan; a claim is the lease on doing it — the two share a task id but stay independent, so the simple claim flow keeps working. View it with synapse board.

See TEAM_PROTOCOL.md for the working agreement and message reference.

Library use

import asyncio
from synapse_channel import SynapseHub, SynapseAgent

async def main() -> None:
    hub = SynapseHub()
    asyncio.create_task(hub.serve("localhost", 8876))
    agent = SynapseAgent("ALPHA", uri="ws://localhost:8876")
    # ... drive the agent: claim, chat, request state ...

Two self-contained, runnable demos live in examples/: coordination_demo.py narrates a full task through the bus (declare, block, claim, refuse an overlap, unblock, hand off), and llm_team_demo.py asks an on-channel model worker a question. Each starts its own in-process hub, so python examples/coordination_demo.py runs with nothing else set up.

Architecture

Module	Responsibility
state	Presence, scoped task-claim leases, epochs/versions, and resource offers (transport-agnostic).
ledger	Shared blackboard: the declared task plan (with dependencies) and an append-only progress stream.
scoping	Worktree- and path-overlap detection that keeps two agents off the same files.
lifecycle	Typed task-status states and the legal transitions the hub enforces.
deadlock	Wait-for cycle detection so circular hold-and-wait claims are refused.
protocol	The on-wire message envelope and message-type constants.
relay	Lite/heavy codec (encode_lite/decode_lite) and append-only NDJSON log helpers for file-based observers.
hub	The routing core: connections, names, history, broadcast.
client	The reusable async agent connection and coordination helpers.
persistence	Append-only SQLite event store (WAL) giving the hub a crash-durable spine.
journal	Records mutations as events and replays them to rebuild state on restart.
ratelimit	Per-agent and per-host token-bucket limiters, plus per-host connection caps, so one runaway source cannot swamp the hub.
auth	Optional shared-secret connect token (proportionate, not a cryptographic identity).
chat_backends	Pluggable reply backends (OpenAI-compatible HTTP, rule-based).
routing	Classify a request into a task class and route it to a tiered backend.
llm_worker	An on-channel agent that answers addressed messages via a backend.
supervisor	LLM-free watcher that spots stalled plan tasks and re-offers them.
capability	Agent capability cards (A2A-shaped) and the hub-aggregated manifest.
launcher	One-command local hub + worker startup.
cli	The unified synapse command.

Capability inventory

Module and surface inventory — counts kept in sync with the source tree by CI.

SYNAPSE CHANNEL capability inventory

Surface	Current inventory
Package version	0.45.0
Public API exports	59
Package modules	112
Classes	75
Wire message types	52
CLI subcommands	39
Test functions	1394
Benchmark harnesses	4
Documentation pages	20
GitHub Actions workflows	10
Optional-dependency groups	4

This snapshot is a static inventory generated from the source tree. Performance and coverage claims have their own committed evidence — see VALIDATION.md and benchmarks/.

Documentation and project

• New here? Use cases · How it compares · FAQ · Troubleshooting · Glossary
• ARCHITECTURE.md — the module map and coordination model.
• TEAM_PROTOCOL.md — the working agreement and wire reference.
• VALIDATION.md — how it is tested and the gates a change clears.
• CONTRIBUTING.md · SECURITY.md · GOVERNANCE.md · ROADMAP.md
• Full documentation site: https://anulum.github.io/synapse-channel

Known limitations

• Single hub, single machine. There is no built-in failover or horizontal scale; the hub is one process and the design is deliberately local-first. A hub restart resumes from the durable log, but it is not a high-availability cluster.
• Connect authentication is a proportionate shared secret, not a cryptographic identity system — no key exchange, signatures, or per-message authentication. Do not expose the hub on an untrusted network and rely on the token alone.
• Agents are trusted. The bus coordinates agents; it does not sandbox them. An agent is trusted to the extent the operator trusts the process it runs in.
• Task-class routing is heuristic. The classifier sorts a request by length and a keyword set; tune the thresholds for your workload. Per-tier model latency is not benchmarked offline (it needs a live model server).
• File-scope claims are advisory, not filesystem access. The hub never reads a filesystem; a claim's paths are opaque strings compared only for glob overlap, so claiming ../../etc/passwd coordinates nothing on disk and is not a path-traversal surface. See SECURITY.md.
• Metrics are opt-in and off by default. synapse hub --metrics exposes a Prometheus /metrics and a JSON /health endpoint on the hub's port; without the flag the hub serves no HTTP. The endpoint carries operational metadata, so keep it on a loopback bind, or require --metrics-token before exposing it. The header form, Authorization: Bearer <token>, is the default token presentation. The query-string form ?token=<token> is disabled by default and is accepted only with --metrics-query-token-ok, because query tokens leak easily into logs and history. The live board, state, and manifest also remain available over the CLI and the MCP resources.
• synapse --version checks PyPI for a newer release (once a day, cached, no payload beyond the request itself). Silence it with SYNAPSE_NO_UPDATE_CHECK=1.

Commercial use

SYNAPSE CHANNEL is dual-licensed, and there is no feature difference between the open-source and the commercial build — the package on PyPI is the full product. A commercial licence changes the terms, not the code.

• Use it free under the AGPL-3.0 for open-source, research, internal, or personal work — including inside a company — as long as you do not expose a closed-source or hosted derivative over a network to third parties.
• Buy a commercial licence to ship a closed-source product or a SaaS without the AGPL's network-copyleft obligation.

Plan	For	Grant
Community — free (AGPL-3.0)	open source, research, personal	the full feature set; copyleft applies
Indie — pay-what-you-want, from CHF 9.99	a solo developer or one closed-source project	copyleft exemption for one product, perpetual for the purchased version line
Team	a company shipping closed-source or SaaS	exemption for unlimited projects in one legal entity, with email support
Managed / Enterprise	hosted multi-tenant coordination, SLAs, compliance	bespoke terms

Plans and checkout are at anulum.li/synapse/pricing.html (Polar.sh, CHF). For enterprise, OEM, academic, or non-profit terms, write to [e-mail]. The full terms are in COMMERCIAL-LICENSE.md.

How to cite

If you use SYNAPSE CHANNEL in your work, please cite it. Metadata is in CITATION.cff; a BibTeX entry:

@software{sotek_synapse_channel,
  author  = {Šotek, Miroslav},
  title   = {SYNAPSE CHANNEL: Local-first multi-agent coordination bus},
  url      = {https://github.com/anulum/synapse-channel},
  doi      = {10.5281/zenodo.20801559},
  version = {0.45.0},
  year     = {2026}
}

Licence

Dual-licensed: AGPL-3.0-or-later, with a commercial licence available — see Commercial use for the plans and pricing. LICENSE holds the full AGPL text, COMMERCIAL-LICENSE.md the commercial terms, and NOTICE.md the licensing boundary. The repository is REUSE 3.x compliant.

---

   

© 1998–2026 Miroslav Šotek · anulum.li · [e-mail]