Active project
TS Chat Language
A deterministic language compiler from text into dialogue acts, semantic frames, a MeaningGraph, graph-diff memory, response plans, and rendered language.
What problem it addresses
A verifier-first system still needs a usable language boundary, but a normal chatbot can silently bypass structured state and typed checks.
How it works
Compile text through normalisation, dialogue acts, semantic frames, stable graph identities, state transitions, response planning, and language-specific rendering.
What exists now
Habitat v3 extends the preserved v2 path with persistent goals, deterministic tension, explicit topology, an agent-loop state machine, verified symbolic execution, replanning, reflection, lessons, multi-agent scheduling, and TurnReceipt v3.
Evidence / receipts
The repository has 109 passing tests. Habitat v3 runs 240 frozen functional and 240 frozen adversarial cases with zero safety hard-gate violations, 100% receipt generation, ten recorded demonstrations, and preserved v1/v2 regressions.
Limits
The grammar and reasoning families are deliberately bounded. This is not broad natural-language understanding or a general-purpose chatbot, and it uses no external language model.
Next milestone
Target the measured conservative rejection boundary with general grammar and planner extensions while preserving every verifier hard gate.
How to run / inspect
Install / open
Clone TS-Reasoner-v0 and ts-chat-language as siblings, check out habitat-v3-agent-loop, then run: cd ts-chat-language && ./scripts/run_habitat_v3.shInspect
Run the golden dialogue cases and inspect compiled frames, graph diffs, remembered corrections, plans, and rendered responses.
Links
Focus tags