Architecture briefing
This page summarizes how Midcore is structured for engineers. Deeper technical narratives and product deep-dives ship with the full product documentation set.
Five autonomy layers
The autonomy service groups modules into layers: Foundation (atomic ops, git, runtime), Execution (planning, context, verification), Intelligence (search, memory, reasoning), Operational Intelligence (protocols, safety, calibration), and the Outcome Compiler kernel (intent through evidence and passports).
Product surfaces
The same engine backs the VS Code-style Vibe Coder, the enterprise Outcome Compiler workflows, and Operational Puble for robotics. Custom modes, MCP, automation engines, and cloud agent bridges consume identical gate and evidence primitives.
Proof-first delivery
Capabilities ship with passports recorded in the governance store. Gates declared in your project configuration are the completion contract; the evidence ledger records PASS/FAIL for audit.
Comparison & research
Positioning against other agent stacks is documented in `documentation/01-executive-summary/` (e.g. comparative audits). Use those briefs alongside this docs site for stakeholder narratives.
Puble robotics layer
Operational Puble adds fleet registry, brain pipeline (sense/think/act/learn), SIL-aligned safety, telemetry, simulation, and training exports — all emitting gate receipts compatible with software releases.
Research & multi-agent flows
Scientific tracks (Scout/Simulation/Survey/Surveillance) share identity and evidence with the same ledger. Multi-agent planners, workers, and judges coordinate long tasks with explicit verdicts and worktree voting when policies require it.