MULTI-AGENT COORDINATION

Shared work.
Explicit responsibilities.

Embed policy checks, peer validation, and replayable governance evidence into multi-agent workflows. Build on your existing agent stack with constitutional-swarm.

For teams evaluating governed coding, review, and multi-agent handoffs.

Conceptual coordination flow

Make the handoff inspectable.

Proposer
candidate artifact
Assigned validators
signed peer votes
Quorum settlement
recorded decision
An illustrative mesh flow. Peer identities, signing keys, and persistence need configuration.
Mixed maturity · optional researchSource 1.1.0Python 3.11+AGPL-3.0-or-laterView evidence

WHERE IT FITS

A clear place in your stack.

Use this when multiple agents need declared task dependencies, assigned validators, and evidence of how a shared decision was reached.

  1. 01

    Describe the work

    Define task dependencies and the agents available to claim them.

  2. 02

    Validate with peers

    Assign validators and collect their signed votes through the mesh.

  3. 03

    Settle + retain evidence

    Use the configured quorum and retain settlement records in your selected store.

From one policy check to a coordinated workflow.

AgentDNA
Rules and YAML-backed checks for an individual agent.
TaskDAG
Dependencies, availability, claims, and results.
ConstitutionalMesh
Peer registration, votes, and settlement.
Evidence
Receipt verification and optional persisted settlements.

INSIDE THE PRODUCT

Capabilities you can inspect.

Follow each behavior to its implementation or maintained documentation.

01

Start with each agent’s rules

AgentDNA provides default, rule-based, and YAML configuration, plus validation and a decorator for governed calls.

Inspect the source
02

Represent the handoff

The public API exposes task DAG compilation, task availability, claims, results, and governed commit operations.

Inspect the source
03

Assign validation responsibilities

The mesh supports local and remote peer registration, assigned validators, signed votes, and quorum settlement.

Inspect the source
04

Retain evidence intentionally

Optional JSONL and SQLite settlement stores keep governance evidence. The receipt-verification CLI is part of the declared package interface.

Inspect the source

LOCAL EVALUATION

Begin with a single agent.

The example uses the inspected AgentDNA API. Follow the repository setup before running it; this page does not assert that source version 1.1.0 is published to PyPI.

  1. Prepare the source environment

    Use Python 3.11+ and the repository instructions. The package depends on acgs-lite; review the AGPL license for your use.

  2. Inspect one validation result

    Start with AgentDNA.default and a proposed action. Check the result and constitution hash.

  3. Add coordination deliberately

    Introduce task ownership and peer validation. Configure settlement persistence and retain signing keys before relying on evidence across restarts.

Open full setup instructions Open the pinned checkout
Python · inspected API
from constitutional_swarm import AgentDNA

agent = AgentDNA.default(agent_id="reviewer")
result = agent.validate("Summarize the meeting notes")

print(result.valid)
print(result.constitutional_hash)

One local validation call. The mesh, transport, and task executor are separate integration steps.

Copy the example or select the code manually.

DESIGN AROUND THE BOUNDARY

Know the scope.

The integration and operating conditions are part of the product decision.

Is this a hosted multi-agent platform?

The inspected product is an embeddable Python runtime. This page makes no hosted dashboard, managed transport, or production service claim.

Is evidence persistent by default?

The mesh has no settlement storage configured by default. Choose a persistence backend and retain the signing key if verification must survive process restarts.

Are all integrations equally mature?

No. The README separates core APIs from advanced optional integrations and research modules. LangGraph, Bittensor, and transport are optional; evaluate the surface you intend to use.

What do the current checks establish?

Default Python tests, type checks, and packaging passed at the inspected commit. Overall CI failed on a PostgreSQL outbox contention test timeout. Historic benchmark figures are not current product promises.

SOURCE, RELEASE, AND CHECKS

Evidence with a reference.

1.1.0 is inspected source metadata. Current package publication was not independently verified. The project separates its core API, advanced integrations, and research modules.

INSPECTED SOURCEconstitutional-swarm 1.1.0f40af8dea690

Reviewed 2026-09-08. Source version and published release are tracked separately.

CI SNAPSHOT · 2026-09-08CI needs attention

Default Python 3.11/3.12 tests, type checks, and package checks passed. Overall CI failed when one PostgreSQL authority-store outbox test timed out.

Read the workflow evidence

FROM READING TO EVALUATION

Start with the source.

Follow one example. Inspect the result. Decide whether it fits your workflow.