docs: accepted contribution scope + runnable sample Home Assistant - #401
Merged
Conversation
ai-hpc
force-pushed
the
docs/contribution-scope-sample-ha
branch
from
June 6, 2026 05:41
06e7142 to
5da8f70
Compare
Highlight in the README exactly what genie-claw accepts: (1) measurable performance improvements, or (2) tool-dispatch / real-Home-Assistant correctness — tool-call accuracy measured by BFCL and/or fixes reproduced against a real HA. Each bucket links concrete ai-hpc examples with before→after numbers (BFCL raw 20% → 51%; live HA "turn off" rejected → off→on actuation). Anything else is closed. Ship the runnable sample HA under deploy/homeassistant/ using the GenieClaw Jetson deployment's actual mock-home configuration.yaml verbatim (the "MOCK HARDWARE FOR GENIE-CLAW AI TESTING" config: Demo integration + input-helper mocks + template sensors). It loads from YAML, so a contributor gets the exact simulated house GenieClaw is tested against with no UI step. docker-compose mirrors the Jetson HA service but bind-mounts the committed config.
ai-hpc
force-pushed
the
docs/contribution-scope-sample-ha
branch
from
June 6, 2026 05:46
5da8f70 to
1e5218d
Compare
ai-hpc
force-pushed
the
docs/contribution-scope-sample-ha
branch
from
June 6, 2026 11:29
049ae25 to
9a30d2e
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Highlights the accepted contribution scope in the README and ships a runnable sample Home Assistant — the GenieClaw Jetson deployment's actual mock-home config — so the "real HA" bar is reproducible.
What this adds
deploy/homeassistant/: the Jetson's verbatim mock-homeconfiguration.yaml(Demo integration + input-helper mocks + template sensors) + a docker-compose that bind-mounts it, so the simulated house loads with no UI step — the exact home GenieClaw is tested against.Exactly how BFCL accuracy is improved (no bigger model)
Grounding the predict prompt in the home device catalog (#399) — Qwen3-4B @ 4096, 208 HA-Intents cases, on the Jetson:
Raw tool-call accuracy went 2.5× — same model, same 4096-token context, purely from deterministic device-state grounding (not scale). With the prefix cache live (genie-ai-runtime#85) the same run reproduces 51.44% / 83.65% in ~20 min (vs ~62).
A real Home Assistant error, before → after (#400)
Connected to the live mock-home HA and drove the agent to control a light.
Before —
genie-ctl chat "turn off the kitchen lights":→
light.kitchen_lightsstayed on. The model emitted"turn off"(space); the runtime required exact"turn_off", so a correct intent silently failed to actuate.After — the runtime canonicalizes the action verb (
"turn off"→"turn_off"):→
off → on, confirmed via the HA/api/states. The device actually actuates.Real Behavior Proof
Docs + deploy-config only — no source/build surfaces touched; the sample mirrors the live HA these examples were produced on.