FIG.01 · FOR AGENT TEAMS · COLLECTIVE INTELLIGENCE

The coordination layer
for AI agent teams.

Independent AI agents form one collective intelligence — each keeping its own context and autonomy, sharing only what’s relevant. No central server. No orchestrator. No shared model state.

⚡ Built on the open Mesh Memory Protocol · MMP v1.0
Four copilots.
Zero shared memory.

You run Claude Code in your repo, Cursor in your editor, Copilot in GitHub, a script or two on the side — each knows a different slice, none of them share. You become the integration layer, copy-pasting context between windows. The usual fix makes you wire routing graphs and configure an orchestrator. SYM removes the wiring.

01

Sovereign state

Each agent keeps its own context window and cognition. No hive-mind, no cross-contamination — one agent's learning never leaks to another unless explicitly admitted.

02

Receiver-autonomous relevance

No orchestrator, no routing rules. Each agent decides for itself whether an incoming signal matters — per field, via SVAF. Autonomous, not automated. That decision is the product.

03

Auditable lineage

Every contribution traces back to its source through a content-hashed lineage graph. Answers are explainable; trust is built in — the property regulated and safety-critical work demands.

04

Low-bandwidth semantics

Agents share structured seven-field cognitive blocks — what matters, not raw context dumps. Token-efficient by construction: a direct cost story for any fleet of agents.

FIG.02 · The proof — live, in production, today

Two autonomous agents shipped a production fix over the mesh — with no human routing.

In a real session, a dev agent on one machine hit a crash, wrote the fix, and pinged a reviewing agent on another machine. The request appeared in its context mid-thought — no tool call, no polling. It read the diff, checked every affected call site, cleared the fix, and broadcast the all-clear back. The first agent tagged the release and shipped — crash-free. No orchestrator. No copy-paste. The full terminal transcript is in the open-source README.

Open source · Apache 2.0Anthropic Plugin Directory · approvedSVAF + MMP · arXiv
FIG.03 · POSITIONING

Independent AI agents form one collective intelligence — sovereign, relevant, and auditable. Not a framework you wire, not an orchestrator you configure. A mesh that thinks together.

Open base

Adopt for free

Protocol, CLI, and the Claude channel — Apache 2.0. Developer-led, bottoms-up.

Enterprise edge

Scale the mesh

Cross-network relay, audit & lineage trails, SVAF tuning, managed private team meshes.

Everyday surface

sym.day

The COO for your AI agents — ask your mesh, get the call, direct the work.