Add live chat runtime

chat/__init__.py

@@ -1,3 +1,3 @@
-from .runtime import ChatRuntime
+from .runtime import ChatResponse, ChatRuntime
 
-__all__ = ["ChatRuntime"]
+__all__ = ["ChatResponse", "ChatRuntime"]

chat/__main__.py

@@ -0,0 +1,39 @@
+from __future__ import annotations
+
+try:
+    import readline  # noqa: F401
+except ImportError:  # pragma: no cover - platform optional
+    readline = None
+
+from chat.runtime import ChatRuntime
+
+_DIM = "\033[2m"
+_RESET = "\033[0m"
+
+
+def main() -> None:
+    runtime = ChatRuntime()
+    while True:
+        try:
+            text = input("> ").strip()
+        except EOFError:
+            print()
+            break
+        if text in {"quit", "exit"}:
+            break
+        if not text:
+            continue
+        try:
+            response = runtime.chat(text)
+        except (KeyError, ValueError) as exc:
+            print(f"{_DIM}[{exc}]{_RESET}")
+            continue
+        print(response.surface)
+        print(
+            f"{_DIM}[role={response.dialogue_role} "
+            f"versor_condition={response.versor_condition:.2e}]{_RESET}"
+        )
+
+
+if __name__ == "__main__":
+    main()

chat/runtime.py

@@ -1,25 +1,82 @@
 from __future__ import annotations
 
+from dataclasses import dataclass
 import re
+from collections.abc import Sequence
 
-from language_packs import OOVPolicy, load_pack, load_pack_entries
+import numpy as np
+
+from algebra.versor import versor_condition
+from generate.dialogue import DialogueRole, classify_dialogue_blade, propose_dialogue
+from generate.proposition import FrameRegistry, Proposition, propose
+from generate.stream import generate
+from language_packs import OOVPolicy, load_mounted_packs, load_pack, load_pack_entries
 from persona.motor import PersonaMotor
 from session.context import SessionContext
 
 _TOKEN_RE = re.compile(r"\w+", re.UNICODE)
+_DEFAULT_PACKS = ("en_minimal_v1", "he_logos_micro_v1", "grc_logos_micro_v1")
+_SEED_ALIASES = {
+    "logos": "λόγος",
+    "dabar": "דבר",
+    "or": "אור",
+    "phos": "φῶς",
+    "zoe": "ζωή",
+    "arche": "ἀρχή",
+    "aletheia": "ἀλήθεια",
+}
+
+
+@dataclass(frozen=True, slots=True)
+class ChatResponse:
+    surface: str
+    proposition: Proposition
+    dialogue_role: DialogueRole
+    versor_condition: float
 
 
 class ChatRuntime:
-    def __init__(self, pack_id: str = "en_minimal_v1") -> None:
+    def __init__(
-        manifest, manifold = load_pack(pack_id)
+        self,
-        entries = load_pack_entries(pack_id)
+        pack_id: str | Sequence[str] = _DEFAULT_PACKS,
-        self._manifest = manifest
+        *,
+        frame_pack: str | None = None,
+    ) -> None:
+        pack_ids = (pack_id,) if isinstance(pack_id, str) else tuple(pack_id)
+        manifests = []
+        manifolds = []
+        entries = []
+        for mounted_pack_id in pack_ids:
+            manifest, manifold = load_pack(mounted_pack_id)
+            manifests.append(manifest)
+            manifolds.append(manifold)
+            entries.extend(load_pack_entries(mounted_pack_id))
+
+        manifold = manifolds[0] if len(pack_ids) == 1 else load_mounted_packs(pack_ids)
+        self._manifests = tuple(manifests)
         self._context = SessionContext(manifold, persona=PersonaMotor.identity())
+        self._frame_registry = FrameRegistry.from_pack(
+            frame_pack or self._default_frame_pack(pack_ids),
+            self._context.vocab,
+        )
         self._surface_by_fold = {e.surface.casefold(): e.surface for e in entries}
+        self._surface_by_fold.update(_SEED_ALIASES)
         self._pos_by_surface = {
             e.surface: (e.pos or e.part_of_speech or "X") for e in entries
         }
 
+    @property
+    def session(self) -> SessionContext:
+        return self._context
+
+    @staticmethod
+    def _default_frame_pack(pack_ids: tuple[str, ...]) -> str:
+        if any(pack_id.startswith("grc_") for pack_id in pack_ids):
+            return "grc"
+        if any(pack_id.startswith("he_") for pack_id in pack_ids):
+            return "he"
+        return "en"
+
     def _tokenize(self, text: str) -> list[str]:
         tokens: list[str] = []
         for match in _TOKEN_RE.finditer(text):
@@ -27,6 +84,9 @@ class ChatRuntime:
             tokens.append(self._surface_by_fold.get(raw.casefold(), raw))
         return tokens
 
+    def tokenize(self, text: str) -> list[str]:
+        return self._tokenize(text)
+
     def _apply_oov_policy(self, tokens: list[str]) -> list[str]:
         kept: list[str] = []
         for token in tokens:
@@ -34,9 +94,12 @@ class ChatRuntime:
                 self._context.vocab.get_versor(token)
                 kept.append(token)
             except KeyError:
-                if self._manifest.oov_policy is OOVPolicy.FAIL_CLOSED:
+                if all(manifest.oov_policy is OOVPolicy.FAIL_CLOSED for manifest in self._manifests):
                     raise
-                if self._manifest.oov_policy is OOVPolicy.PROPOSE_VOCAB_EXPANSION:
+                if any(
+                    manifest.oov_policy is OOVPolicy.PROPOSE_VOCAB_EXPANSION
+                    for manifest in self._manifests
+                ):
                     raise KeyError(f"OOV token requires vocab proposal: {token}")
         return kept
 
@@ -51,12 +114,58 @@ class ChatRuntime:
             prev_pos = pos
         return out
 
-    def respond(self, text: str, max_tokens: int = 32) -> str:
+    def _dialogue_reference(self) -> np.ndarray | None:
+        blade = self._context.last_dialogue_blade
+        if blade is None or float(np.linalg.norm(blade)) < 1e-8:
+            return None
+        return blade
+
+    def chat(self, text: str, max_tokens: int = 32) -> ChatResponse:
         tokens = self._tokenize(text)
         filtered = self._apply_oov_policy(tokens)
         if not filtered:
-            return ""
+            raise ValueError("ChatRuntime.chat() received no in-vocabulary tokens.")
-        self._context.ingest(filtered)
+
-        result = self._context.respond(max_tokens=max_tokens)
+        field_state = self._context.ingest(filtered)
+        reference_blade = self._dialogue_reference()
+        base_proposition = propose(field_state, None, self._context.vocab, self._frame_registry)
+        dialogue_role = classify_dialogue_blade(
+            base_proposition.relation,
+            reference_blade,
+        )
+        proposition = propose_dialogue(
+            field_state,
+            None,
+            self._context.vocab,
+            self._frame_registry,
+            reference_blade,
+        )
+        self._context.record_dialogue(proposition)
+
+        result = generate(
+            field_state,
+            self._context.vocab,
+            self._context.persona,
+            max_tokens=max_tokens,
+            vault=None,
+        )
+        self._context.state = result.final_state
+        self._context.vault.store(
+            result.final_state.F,
+            {"turn": self._context.turn, "role": "assistant"},
+        )
+        self._context.turn += 1
         guarded = self._syntactic_guard(result.tokens)
-        return " ".join(guarded)
+        surface = " ".join(guarded)
+        return ChatResponse(
+            surface=surface,
+            proposition=proposition,
+            dialogue_role=dialogue_role,
+            versor_condition=versor_condition(result.final_state.F),
+        )
+
+    def respond(self, text: str, max_tokens: int = 32) -> str:
+        try:
+            return self.chat(text, max_tokens=max_tokens).surface
+        except ValueError:
+            return ""

probe/repl.py

@@ -1,30 +1,20 @@
 from __future__ import annotations
 
-from language_packs import load_mounted_packs
+from chat.runtime import ChatRuntime
-
-_TRILINGUAL_PACKS = ("en_minimal_v1", "he_logos_micro_v1", "grc_logos_micro_v1")
-_SEED_ALIASES = {
-    "logos": "λόγος",
-    "dabar": "דבר",
-    "or": "אור",
-    "phos": "φῶς",
-    "zoe": "ζωή",
-    "arche": "ἀρχή",
-    "aletheia": "ἀλήθεια",
-}
 
 
 def field_walk(seed: str, steps: int = 4) -> list[str]:
-    vocab = load_mounted_packs(_TRILINGUAL_PACKS)
+    runtime = ChatRuntime()
-    surface = _SEED_ALIASES.get(seed.casefold(), seed)
+    injected_terms = runtime.tokenize(seed)
-    F = vocab.get_versor(surface)
+    response = runtime.chat(seed, max_tokens=max(1, steps))
-    walk = [seed] if surface == seed else [seed, surface]
+    proposition_terms = [
-    idx = vocab.index_of(surface)
+        response.proposition.subject,
-    for _ in range(max(0, steps - len(walk))):
+        response.proposition.predicate,
-        word, idx = vocab.nearest(F, exclude_idx=idx)
+    ]
-        walk.append(word)
+    if response.proposition.object_ is not None:
-        F = vocab.get_versor(word)
+        proposition_terms.append(response.proposition.object_)
-    return walk
+    walk = [seed, *injected_terms, *proposition_terms, *response.surface.split()]
+    return walk[: max(1, steps)]
 
 
 def main() -> None:
@@ -37,7 +27,7 @@ def main() -> None:
             break
         try:
             chain = field_walk(text)
-            print(f"[field walk: {' → '.join(chain)}]")
+            print(f"[field walk: {' -> '.join(chain)}]")
         except KeyError:
             print("[unknown token]")
 

tests/test_chat_runtime.py

@@ -0,0 +1,23 @@
+from __future__ import annotations
+
+import numpy as np
+
+from chat.runtime import ChatResponse, ChatRuntime
+
+
+def test_chat_runtime_keeps_live_session_across_two_turns() -> None:
+    runtime = ChatRuntime()
+
+    first = runtime.chat("light logos", max_tokens=8)
+    first_field = runtime.session.state.F.copy()
+
+    second = runtime.chat("light truth", max_tokens=8)
+    second_field = runtime.session.state.F.copy()
+
+    assert isinstance(first, ChatResponse)
+    assert first.surface.strip()
+    assert second.surface.strip()
+    assert first.versor_condition < 1e-6
+    assert second.versor_condition < 1e-6
+    assert second.dialogue_role in {"elaborate", "assert"}
+    assert not np.array_equal(second_field, first_field)