[Pentexa]

METARGON NEST v0.1 — the first interactive home

A local browser application where the founder can talk to METARGON, assign real challenges, watch it explore, inspect its learning, and save/restore progress. The intelligence is the real Academy student agent (MetargonAcademyAgent) + MechanismGrid; no scripted agent, no LLM decisions.

Highlights

• Real agent loop (observe -> plan -> act -> learn) mirroring the Academy experiment harness; hidden rules never reach the agent.
• Deterministic command interpreter (works with no API key) covering greetings, memory questions, real goals ("open door D1/D2", "reach the goal"), controls, persistence, and "why did you do that?" from telemetry.
• Optional OpenAI-compatible adapter (user key) that only translates language to the same structured intents — never decides actions.
• Live world canvas (agent, doors, switches, plates, objects, goal, last action, outcomes) with Start/Pause/Continue/Single-step/Stop/Reset.
• Learning & memory panel from real telemetry: hypotheses + status, support/contradiction, confidence, cross-episode flags, trace counters.
• Persistence: agent's own save_state/load_state into an application-owned gitignored /.nest/ directory; restore is exact; incompatible checkpoints are refused.
• Honesty: unsupported requests get honest refusals; unverified hypotheses are presented as such; nothing is invented.
• Safety: localhost-only, bounded execution (episode+global caps), no shell execution, cancellation-safe worker.
• CI job runs the NEST integration tests (seconds).

Demos (automated evidence in

est/demo_evidence.txt)

1. First encounter — greeting + honest "no memory yet".
2. First task (open door D1) — 80 real steps, door D1 opened, 74 plans used.
3. Learning — 10 hypotheses generated, 10 verified (e.g. "door D1 opens when empty(P1)").
4. Persistence — save->restore identical memory (10 vs 10 hypotheses).
5. Honest failure — fly/execute/give-me-emotions all refused, never claimed.
6. No API — full workflow with api_mode=none.

Tests

est/tests/test_nest.py (8 tests): the six demos + server endpoints.

• Audit suite 170/170 OK; Academy subset (gridworld/agents/memory/memory_v06) 54/54 OK; repo hygiene OK.

Files

est/ — app (session, interpreter, server, storage, static UI, launch, demo/tests).

• docs/METARGON_NEST.md — launch & usage.
• CI + .gitignore updated.

Launch: python nest/launch.py from the repo root.

No merge performed; please review.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.