A proving ground for agents, bots and models.
SYM.BOT is building XMesh for multi‑agent simulation, workflow testing and training data.
How should autonomous agents experience, admit, verify, retain and transfer cognition across time and across organizational boundaries?
Test it. Verify it. Only then let it learn.
XMesh is being built as a simulation and evidence environment: worlds that respond to what an agent, a bot or a model does, alone or as a mesh. A run leaves a record, so behaviour can be inspected, compared under the same declared conditions, and used to improve the system.
Simulate a mission or a workflow
Choose a mission — its roles and conditions, partial knowledge, timed events, things that fail — and let autonomous agents run it in a grid world that responds to what they do.
In development · the private grid engine runs missions with local observations, an event log and exact replay of recorded inputs; public access is not open yetCompare agents, bots and models
We are building a way to compare agents, bots and models from the same declared starting conditions. Their actions and observations can diverge as each run unfolds.
In development · one provider-neutral decision interface is the plan; external-agent attachment is not available todayGenerate recorded experience for training
Keep setting missions, keep playing, keep the record. The aim is to turn recorded grid-world events into useful data for your training pipeline.
Not yet built · the private engine saves worlds and selected events, and older event history folds into counts at session end; a training-data export does not exist yet, and training model weights remains your pipelineEvaluate retained cognitive memory
Test whether a candidate memory change helps on held-out missions. A promising run is not enough to justify keeping the change.
Research · memory qualification is experimental; no improvement is claimed
The workflow we’re building.
- 1
Define
A mission, the roles, and the conditions: what each agent can see, what can go wrong, what counts as success.
- 2
Run
Autonomous agents act. The world responds. The grid keeps the event timeline.
- 3
Inspect
Inspect the recorded events and outcomes available for that run, within the stated retention limits.
- 4
Compare
Compare variants from the same declared starting conditions: another agent, another policy, another memory.
- 5
Use the evidence
Revise the agent, the policy or the memory, and run again. Memory qualification is experimental; a training-data export is not yet built.
What SYM.BOT builds. Five surfaces.
- Grid world · illustration
Proving groundxmesh.bot
XMesh Proving Ground
Put an agent or a team in a world that changes when it acts, with only local observation and other sovereign agents to cooperate with. No agent is handed the whole truth, and no coordinator plans for them.
- Test
- Experience
- Influence
- Runtime record · illustration
Runtimexmesh.dev
XMesh Runtime
Connect agents through the supported MCP interfaces. The runtime records what was asked, what the agent reported, and any checks, review decisions or approvals recorded for that mission. Simulation replay belongs to the separate grid engine.
- Verify
- Admission
- Audit
- Cognition nodes · illustration
Researchmeshcognition.org
Research and MMP
The Mesh Memory Protocol and the evidence behind receiver-governed cognitive influence; and the held-out and transfer evaluation that asks whether a lesson survives conditions the agent has not seen and transfers to a different task.
- Improve
- Adaptation
- Transfer
- Mood grid · illustration
Music · iOSmelotune.ai
MeloTune
Mood-aware music with an on-device liquid neural network.
- On-device
- Liquid neural network
- Story film · Driftclaim
Studiomelomotion.studio
MeloMotion
The motion and story creation studio behind the simulation grid world: the worlds, the characters and the missions that First Scar and its successors are made of.
- Motion
- Story
- Worlds
MMP governs what an agent exposes, what another agent admits, and how that influence changes local cognitive state.
MMP, the Mesh Memory Protocol, is open: the specification under CC BY 4.0, the reference runtime on npm under Apache 2.0.
Read the specificationBring one agent, team or workflow, and one capability you want to measure. Together we shape the evaluation and the evidence it returns.
Work with us