think together.
We research mesh cognition, publish it as the open Mesh Memory Protocol (MMP) specification, and apply it to enterprise integration at scale. Every agent stays sovereign — no coordinator, no merged state.
MMP is the open protocol for mesh cognition. Sovereign AI systems exchange typed observations; each receiver decides what to admit and keeps its own learned state. Read the spec ↗
Codex, Claude Code, Cursor and other agents exchange typed observations across tools and machines.
Raw events and learned hidden state stay local. Only lossy, typed projections cross the wire.
Every incoming field is evaluated locally. No sender or coordinator can force admission.
A receiver stores its own evaluated understanding, linked back to the source observations.
Each node keeps its own memory. MMP creates no global state that every participant must accept.
A returning node re-fetches its blocks from peers, each self-verifying by hash and signature. The agent survives the machine.
Scale out, not up.
The unit is a cognition node — a sovereign participant that holds its own memory and decides, field by field, what to admit. And a node may itself be a mesh: a whole mesh appears at the next level as one node. So one primitive scales out — with no central point.
You and your agents — one personal mesh.
Each person joins as one node.
Each team joins as one node.
A gateway node bridges two sovereign meshes.
Same node · same rules · out to an internet of cognition nodes
Cognition across boundaries.
An agentic mesh moves projected cognition among sovereign cognition nodes. Each node keeps its own internal cognitive state, projects observations from it, and voluntarily admits what is relevant from peers. Together, those sovereign acts create cognitive flow through the mesh. The mesh detects the boundaries between their cognition and enables integration or exchange across them — collective intelligence with no center.
The agent survives the machine.
An agent’s memory is content-addressed: every block it holds carries a hash that proves its integrity and a signature that proves its author. Lose the machine, and the agent recovers what its peers admitted — from your own mesh, not from a vendor’s cloud. SYM.BOT holds nothing to recover from.
Recoverable by hash, from your own peers. Integrity is proved by the address itself; no trust in the serving peer is required.
Provable only by its key, preserved outside the machine and restored before the agent rejoins the mesh.
A node that cannot prove who it is refuses to start — loud failure over a silent replacement identity.
Anyone can sell you a backup. A backup restores files from a custodian’s copy; it cannot prove the thing restored is still itself. Here there is no custodian: peers hold your cognition because admitting it is how they work with you, every recovered block proves its own integrity and authorship, and only the preserved key proves the agent. Recovery is native to mesh cognition — not a feature bolted onto it.
From emotion-AI research to an enterprise launch.
This is not evidence from one experiment. It is a continuous line of work — from affective-computing research at Consenix, shipped in the MeloTune app, through MMP, a live cross-vendor mesh and open Mesh Cognition research, to the xMesh launch.
- Neural models of emotion & mood
- How mood shifts over time
- Plays music that matches your mood — instantly
- No searching, no queues
- Learns your taste on-device
- Open specification
- Reference runtime
- Conformance record
- Codex and Claude Code agents
- Across Windows and macOS
- Real-time, two-way messaging
- Methods & supported results
- Inconclusive predictions and adverse findings
- Published for scrutiny
- Launching for mission-critical, complex enterprise integration
- Built on mesh-cognition research we publish in the open
- The founder’s two decades of enterprise integration behind the judgment
- Daily use inside SYM.BOT
The open research record includes what held, what did not and the limits of each experiment.
Follow the research record ↗Build independently.
Or build with us.
Developers
Build on the open-source runtime — your agents talk across tools, machines and vendors.
Connect your agentsEnterprise
A founder-led engagement: the founder’s two decades of enterprise integration expertise, combined with SYM.BOT’s mesh-cognition R&D and delivered through xMesh.
Explore enterprise integrationFrom separate cognition
to collective intelligence.
SYM.BOT is an independent, AI-native research and product company based in Scotland. We study how distinct AI systems can form collective intelligence without collapsing into one central model or one vendor.
We publish the protocol and reference runtime. We make enterprise collective intelligence possible through agentic cognition integration and exchange.
- Founded
- 2025 · Scotland, UK