ArcNet [WIP]

Maintenance layer for agent fleets: watch, fix, prove

[github]

// what it does

ArcNet watches agent fleets: behavior, cost, and the trust of everything an agent ingests. Untrusted inputs (scraped pages, tool outputs) get trust-tagged by Unplug before they reach the model. When something slips through, ArcNet traces it with OpenTelemetry into SigNoz, alerts on it, and signals the agent to self-correct via Agno guardrails and run cancellation.

// the loop

agent runs -> OTel telemetry -> SigNoz -> alert fires -> webhook -> ArcNet signal
     ^                                                              |
     +------ agent pauses / self-corrects / is quarantined <--------+

// the time machine

Replay a recorded incident against a different model or prompt, with the same guardrails, and prove it would behave better: goal reached, fewer steps, lower cost, attack resisted. Tool outputs are mocked from the transcript, so the replay is faithful. Trace history becomes a behavioral regression suite - the answer to "can we upgrade the model?" that isn't swap-and-pray.

// agent-view

Every datum has a machine-optimal twin (GET /api/agent-view/{view}/{id}), so the coding agents you already run can read fleet health, signals, and incidents in their own format. Agents that watch themselves, and get better.

// stack

Python, OpenTelemetry -> SigNoz, Unplug trust tagging, Agno guardrails, Time Machine replay engine, REST API.

// what's next