prompture 1.11.2.dev3__tar.gz → 1.12.1.dev1__tar.gz

{prompture-1.11.2.dev3 → prompture-1.12.1.dev1}/.env.copy

@@ -162,6 +162,22 @@ COHERE_EMBEDDING_MODEL=embed-v4.0
 VOYAGE_EMBEDDING_MODEL=voyage-3.5
 JINA_EMBEDDING_MODEL=jina-embeddings-v3
 
+# ─── Decision models ("System One": typed decisions, not text) ──────────────
+# TypeSafe (hosted Jev) — key from https://console.typesafe.ai/
+TYPESAFE_API_KEY=
+TYPESAFE_BASE_URL=https://api.typesafe.ai
+TYPESAFE_DECISION_MODEL=jev-latest
+
+# Kev — self-hosted, serves the same /v1/systemone contract. No key needed.
+#   uv run --extra serve python -m kev.serve --run jaredpalmer/kev-4b --port 8009
+KEV_BASE_URL=http://127.0.0.1:8009
+KEV_API_KEY=
+KEV_DECISION_MODEL=kev-latest
+
+# Laya — in-process open weights, no endpoint. Install: pip install prompture[laya]
+LAYA_DECISION_MODEL=router
+LAYA_DEVICE=
+
 # ─── Phase 4: AWS Bedrock ───────────────────────────────────────────────────
 # boto3's standard credential chain is also honoured (IAM role, ~/.aws/credentials).
 # Install: pip install prompture[bedrock]

{prompture-1.11.2.dev3 → prompture-1.12.1.dev1}/.github/workflows/documentation.yml

@@ -1,6 +1,7 @@
 name: documentation
 
 on:
+  workflow_dispatch:
   workflow_run:
     workflows: ["Publish to PyPI and GitHub Release"]
     types: [completed]
@@ -11,7 +12,7 @@ permissions:
 jobs:
   docs:
     runs-on: ubuntu-latest
-    if: ${{ github.event.workflow_run.conclusion == 'success' }}
+    if: ${{ github.event_name == 'workflow_dispatch' || github.event.workflow_run.conclusion == 'success' }}
     steps:
       - uses: actions/checkout@v4
         with:
@@ -26,8 +27,9 @@ jobs:
         run: |
           cacaodocs build ./prompture -o ./docs-cacao -c ./cacao.yaml
       - name: Export static site
+        working-directory: docs-cacao
         run: |
-          cacao build ./docs-cacao/app.py -o ./_build
+          cacao build app.py -o ../_build
       - name: Deploy to GitHub Pages
         uses: peaceiris/actions-gh-pages@v3
         with:

{prompture-1.11.2.dev3 → prompture-1.12.1.dev1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: prompture
-Version: 1.11.2.dev3
+Version: 1.12.1.dev1
 Summary: Ask LLMs to return structured JSON and run cross-model tests. API-first.
 Author-email: Juan Denis <juan@vene.co>
 License-Expression: MIT
@@ -49,6 +49,8 @@ Requires-Dist: ruff>=0.8.0; extra == "dev"
 Requires-Dist: prompture[all]; extra == "dev"
 Provides-Extra: airllm
 Requires-Dist: airllm>=2.8.0; extra == "airllm"
+Provides-Extra: laya
+Requires-Dist: laya>=0.3.6; extra == "laya"
 Provides-Extra: bedrock
 Requires-Dist: boto3>=1.34.0; extra == "bedrock"
 Provides-Extra: ingest
@@ -159,6 +161,7 @@ print(person.name)  # Maria
 **Providers & modalities**
 - 36+ providers under a unified `provider/model` string — see [Providers](#providers)
 - Multi-modal drivers for embeddings, rerank, moderation, image, video, TTS, STT, and audio transforms — see [Multi-Modal](#multi-modal)
+- **Decision models** ("System One") — typed `noul`/`choice`/`score` answers with calibrated probabilities instead of generated text. Hosted (TypeSafe Jev), self-hosted (Kev), or in-process (Laya) behind one interface — see [Decision](#decision)
 - TOON input conversion for 45–60% token savings on structured input ([python-toon](https://github.com/jhd3197/python-toon))
 
 **Agents, tools, RAG**
@@ -339,6 +342,9 @@ Model strings use `"provider/model"` format. The provider prefix routes to the c
 | `jina` | `jina/jina-embeddings-v3` (embedding), `jina/jina-reranker-v2-base-multilingual` (rerank) | Automatic |
 | `nomic` | `nomic/nomic-embed-text-v1.5` (embedding) | Automatic |
 | `mixedbread` | `mixedbread/mxbai-embed-large-v1` (embedding), `mixedbread/mxbai-rerank-large-v1` (rerank) | Automatic |
+| `typesafe` | `typesafe/jev-latest` (decision) | Automatic |
+| `kev` | `kev/kev-4b` (decision, self-hosted) | Free |
+| `laya` | `laya/router` (decision, in-process) | Free (local) |
 | `openai_compatible` | `openai_compatible/<profile>/<model>` — 9 curated profiles: `fireworks`, `together`, `cerebras`, `sambanova`, `perplexity`, `nvidia`, `deepinfra`, `siliconflow`, `github_models` (or pass an explicit `endpoint=` for anything else) | Automatic where pricing is known |
 
 </details>
@@ -380,6 +386,7 @@ Beyond text LLMs, Prompture exposes drivers for adjacent modalities under the sa
 - **Embeddings** — OpenAI (`text-embedding-3-*`), Cohere (`embed-v4.0`), Voyage AI (`voyage-3.5`, `voyage-3-large`), Jina AI (`jina-embeddings-v3`), Nomic (`nomic-embed-text-v1.5`), Mixedbread (`mxbai-embed-large-v1`, `mxbai-embed-2d-large-v1`), and Ollama (`nomic-embed-text`)
 - **Rerank** — Cohere (`rerank-v3.5`), Voyage AI (`rerank-2.5`), Jina AI (`jina-reranker-v2-base-multilingual`), Mixedbread (`mxbai-rerank-large-v1`, `mxbai-rerank-base-v1`, `mxbai-rerank-xsmall-v1`)
 - **Moderation** — OpenAI (`omni-moderation-latest` — free multimodal), Mistral (`mistral-moderation-latest`)
+- **Decision** ("System One" typed decisions) — TypeSafe (`jev-latest` — hosted), Kev (`kev-0.8b`/`kev-4b`/`kev-9b` — self-hosted, same endpoint), Laya (`router`, `english`, `multilingual`, `typed-decisions` — in-process)
 - **Image generation** — OpenAI DALL-E + GPT image, Google Imagen, Grok, Stability AI, Runway (`gen4_image`, `gen4_image_turbo`, `gpt_image_2`, `gemini_image3_pro`, `gemini_2.5_flash`), Kling AI, Fal.ai, Ideogram (v3 — strong typography), Black Forest Labs / Flux (`flux-pro-1.1`, `flux-pro-1.1-ultra`, `flux-dev`, `flux-schnell`, `flux-kontext-pro`/`max` for editing)
 - **Video generation** — Grok Imagine Video; Runway text/image/video → video (`gen4.5`, `gen4_turbo`, `gen3a_turbo`, `gen4_aleph`, `veo3`, `veo3.1`, `veo3.1_fast`); MiniMax / Hailuo; Kling AI; Luma AI Dream Machine (`ray-2`, `ray-flash-2`, `ray-1-6`); Pika Labs (`pika-2.2`, `pika-2.1`, `pika-1.5`); Fal.ai
 - **Text-to-speech** — OpenAI (`tts-1`), ElevenLabs, Cartesia (`sonic-2`), Deepgram (`aura-2-thalia-en`), Runway (`eleven_multilingual_v2`)
@@ -463,6 +470,59 @@ for r in results:
 
 OpenAI moderation is free of charge (`cost == 0`, `pricing_unknown == False`). Mistral moderation is billed at ~$0.10 per million input tokens. Discover configured moderation models with `get_available_moderation_models()`. The async factory is `get_async_moderation_driver_for_model()`.
 
+### Decision
+
+Decision models ("System One") answer typed questions about a state instead of generating text. You send a `state` plus a map of questions; you get back one typed answer per question with calibrated probabilities — nothing to parse, nothing to hallucinate.
+
+Three primitives: `Noul` (yes/no → probability), `Choice` (pick one → winner + distribution + confidence), and `Score` (ordinal rubric → weighted value + confidence).
+
+```python
+from prompture.drivers.decision_base import Choice, Noul, Score
+from prompture.drivers.decision_registry import get_decision_driver_for_model
+
+driver = get_decision_driver_for_model("typesafe/jev-latest")
+result = driver.decide(
+    state={"subject": "Duplicate charge", "body": "We were billed twice. Refund today or we cancel."},
+    questions={
+        "department": Choice(
+            "Which team should handle this?",
+            criteria={"billing": "invoices, refunds", "technical": "bugs, outages", "sales": None},
+        ),
+        "urgency": Score("How urgent is this?", criteria=["not urgent", "this week", "blocking"]),
+        "churn_risk": Noul("Does the sender threaten to cancel?"),
+    },
+)
+
+print(result["department"].choice, result["department"].confidence)  # billing 0.81
+print(result["urgency"].score, result["urgency"].nearest_level)      # 1.05 this week
+print(result["churn_risk"].noul)                                     # 0.95
+print(driver.last_usage["cost"])
+```
+
+All questions in one call share a single reading of the state, so batching beats looping. Because the probabilities are calibrated, the common use is a confidence gate in front of an expensive model:
+
+```python
+answer = result["department"]
+if answer.confidence >= 0.85:
+    route_automatically(answer.choice)   # no LLM call at all
+else:
+    escalate_to_llm(state)
+```
+
+The same question map works across every provider — only the model string changes:
+
+| Model string | Where it runs | Cost | Setup |
+| --- | --- | --- | --- |
+| `typesafe/jev-latest` | hosted | $0.042 / 1M input tokens, output free | `TYPESAFE_API_KEY` |
+| `kev/kev-4b` | your own server, same `/v1/systemone` contract | free | `KEV_BASE_URL` (no key) |
+| `laya/router` | in-process weights | free | `pip install prompture[laya]` |
+
+The open checkpoints trade accuracy for cost and both degrade on high-cardinality choices (50+ options), so benchmark on your own labels before switching a workload off the hosted model.
+
+Discover configured decision models with `get_available_decision_models()` — TypeSafe appears when its key is set, Kev when `KEV_BASE_URL` points at a server, Laya when the package is installed. The async factory is `get_async_decision_driver_for_model()`.
+
+Runnable example: `python examples/decision_model_example.py`.
+
 ### Runway
 
 Runway is a single API surface covering image, video, and audio. One key (`RUNWAY_API_KEY`, or `RUNWAYML_API_SECRET`) unlocks all of it:

{prompture-1.11.2.dev3 → prompture-1.12.1.dev1}/README.md

@@ -54,6 +54,7 @@ print(person.name)  # Maria
 **Providers & modalities**
 - 36+ providers under a unified `provider/model` string — see [Providers](#providers)
 - Multi-modal drivers for embeddings, rerank, moderation, image, video, TTS, STT, and audio transforms — see [Multi-Modal](#multi-modal)
+- **Decision models** ("System One") — typed `noul`/`choice`/`score` answers with calibrated probabilities instead of generated text. Hosted (TypeSafe Jev), self-hosted (Kev), or in-process (Laya) behind one interface — see [Decision](#decision)
 - TOON input conversion for 45–60% token savings on structured input ([python-toon](https://github.com/jhd3197/python-toon))
 
 **Agents, tools, RAG**
@@ -234,6 +235,9 @@ Model strings use `"provider/model"` format. The provider prefix routes to the c
 | `jina` | `jina/jina-embeddings-v3` (embedding), `jina/jina-reranker-v2-base-multilingual` (rerank) | Automatic |
 | `nomic` | `nomic/nomic-embed-text-v1.5` (embedding) | Automatic |
 | `mixedbread` | `mixedbread/mxbai-embed-large-v1` (embedding), `mixedbread/mxbai-rerank-large-v1` (rerank) | Automatic |
+| `typesafe` | `typesafe/jev-latest` (decision) | Automatic |
+| `kev` | `kev/kev-4b` (decision, self-hosted) | Free |
+| `laya` | `laya/router` (decision, in-process) | Free (local) |
 | `openai_compatible` | `openai_compatible/<profile>/<model>` — 9 curated profiles: `fireworks`, `together`, `cerebras`, `sambanova`, `perplexity`, `nvidia`, `deepinfra`, `siliconflow`, `github_models` (or pass an explicit `endpoint=` for anything else) | Automatic where pricing is known |
 
 </details>
@@ -275,6 +279,7 @@ Beyond text LLMs, Prompture exposes drivers for adjacent modalities under the sa
 - **Embeddings** — OpenAI (`text-embedding-3-*`), Cohere (`embed-v4.0`), Voyage AI (`voyage-3.5`, `voyage-3-large`), Jina AI (`jina-embeddings-v3`), Nomic (`nomic-embed-text-v1.5`), Mixedbread (`mxbai-embed-large-v1`, `mxbai-embed-2d-large-v1`), and Ollama (`nomic-embed-text`)
 - **Rerank** — Cohere (`rerank-v3.5`), Voyage AI (`rerank-2.5`), Jina AI (`jina-reranker-v2-base-multilingual`), Mixedbread (`mxbai-rerank-large-v1`, `mxbai-rerank-base-v1`, `mxbai-rerank-xsmall-v1`)
 - **Moderation** — OpenAI (`omni-moderation-latest` — free multimodal), Mistral (`mistral-moderation-latest`)
+- **Decision** ("System One" typed decisions) — TypeSafe (`jev-latest` — hosted), Kev (`kev-0.8b`/`kev-4b`/`kev-9b` — self-hosted, same endpoint), Laya (`router`, `english`, `multilingual`, `typed-decisions` — in-process)
 - **Image generation** — OpenAI DALL-E + GPT image, Google Imagen, Grok, Stability AI, Runway (`gen4_image`, `gen4_image_turbo`, `gpt_image_2`, `gemini_image3_pro`, `gemini_2.5_flash`), Kling AI, Fal.ai, Ideogram (v3 — strong typography), Black Forest Labs / Flux (`flux-pro-1.1`, `flux-pro-1.1-ultra`, `flux-dev`, `flux-schnell`, `flux-kontext-pro`/`max` for editing)
 - **Video generation** — Grok Imagine Video; Runway text/image/video → video (`gen4.5`, `gen4_turbo`, `gen3a_turbo`, `gen4_aleph`, `veo3`, `veo3.1`, `veo3.1_fast`); MiniMax / Hailuo; Kling AI; Luma AI Dream Machine (`ray-2`, `ray-flash-2`, `ray-1-6`); Pika Labs (`pika-2.2`, `pika-2.1`, `pika-1.5`); Fal.ai
 - **Text-to-speech** — OpenAI (`tts-1`), ElevenLabs, Cartesia (`sonic-2`), Deepgram (`aura-2-thalia-en`), Runway (`eleven_multilingual_v2`)
@@ -358,6 +363,59 @@ for r in results:
 
 OpenAI moderation is free of charge (`cost == 0`, `pricing_unknown == False`). Mistral moderation is billed at ~$0.10 per million input tokens. Discover configured moderation models with `get_available_moderation_models()`. The async factory is `get_async_moderation_driver_for_model()`.
 
+### Decision
+
+Decision models ("System One") answer typed questions about a state instead of generating text. You send a `state` plus a map of questions; you get back one typed answer per question with calibrated probabilities — nothing to parse, nothing to hallucinate.
+
+Three primitives: `Noul` (yes/no → probability), `Choice` (pick one → winner + distribution + confidence), and `Score` (ordinal rubric → weighted value + confidence).
+
+```python
+from prompture.drivers.decision_base import Choice, Noul, Score
+from prompture.drivers.decision_registry import get_decision_driver_for_model
+
+driver = get_decision_driver_for_model("typesafe/jev-latest")
+result = driver.decide(
+    state={"subject": "Duplicate charge", "body": "We were billed twice. Refund today or we cancel."},
+    questions={
+        "department": Choice(
+            "Which team should handle this?",
+            criteria={"billing": "invoices, refunds", "technical": "bugs, outages", "sales": None},
+        ),
+        "urgency": Score("How urgent is this?", criteria=["not urgent", "this week", "blocking"]),
+        "churn_risk": Noul("Does the sender threaten to cancel?"),
+    },
+)
+
+print(result["department"].choice, result["department"].confidence)  # billing 0.81
+print(result["urgency"].score, result["urgency"].nearest_level)      # 1.05 this week
+print(result["churn_risk"].noul)                                     # 0.95
+print(driver.last_usage["cost"])
+```
+
+All questions in one call share a single reading of the state, so batching beats looping. Because the probabilities are calibrated, the common use is a confidence gate in front of an expensive model:
+
+```python
+answer = result["department"]
+if answer.confidence >= 0.85:
+    route_automatically(answer.choice)   # no LLM call at all
+else:
+    escalate_to_llm(state)
+```
+
+The same question map works across every provider — only the model string changes:
+
+| Model string | Where it runs | Cost | Setup |
+| --- | --- | --- | --- |
+| `typesafe/jev-latest` | hosted | $0.042 / 1M input tokens, output free | `TYPESAFE_API_KEY` |
+| `kev/kev-4b` | your own server, same `/v1/systemone` contract | free | `KEV_BASE_URL` (no key) |
+| `laya/router` | in-process weights | free | `pip install prompture[laya]` |
+
+The open checkpoints trade accuracy for cost and both degrade on high-cardinality choices (50+ options), so benchmark on your own labels before switching a workload off the hosted model.
+
+Discover configured decision models with `get_available_decision_models()` — TypeSafe appears when its key is set, Kev when `KEV_BASE_URL` points at a server, Laya when the package is installed. The async factory is `get_async_decision_driver_for_model()`.
+
+Runnable example: `python examples/decision_model_example.py`.
+
 ### Runway
 
 Runway is a single API surface covering image, video, and audio. One key (`RUNWAY_API_KEY`, or `RUNWAYML_API_SECRET`) unlocks all of it:

prompture-1.12.1.dev1/VERSION

@@ -0,0 +1 @@
+1.12.0

{prompture-1.11.2.dev3 → prompture-1.12.1.dev1}/prompture/__init__.py

@@ -1,6 +1,8 @@
 """prompture - API package to convert LLM outputs into JSON + test harness."""
 
-from dotenv import load_dotenv
+import os as _os
+
+from dotenv import dotenv_values, find_dotenv
 
 from .agents import *
 
@@ -231,8 +233,35 @@ try:
 except ImportError:
     pass
 
+
+def _load_dotenv_skipping_blanks(path: str | None = None) -> None:
+    """Load ``.env`` into ``os.environ``, ignoring keys with a blank value.
+
+    A placeholder line like ``HF_ENDPOINT=`` means "not configured", but
+    exporting it makes the variable *set-but-empty*, which is not the same
+    thing to the rest of the process. Libraries that resolve their default
+    with ``os.getenv(KEY, default)`` — huggingface_hub's endpoint is one —
+    take the empty string over their default and break, even though the
+    caller never configured anything. Since ``.env`` values never override
+    real environment variables here, a blank placeholder can only ever
+    create such a variable, so there is nothing to lose by skipping it.
+
+    Args:
+        path: Explicit ``.env`` to read. When omitted, the file is located
+            the same way :func:`dotenv.load_dotenv` would — by walking up
+            from this module.
+    """
+    path = path or find_dotenv()
+    if not path:
+        return
+    for key, value in dotenv_values(path).items():
+        if value is None or not value.strip():
+            continue
+        _os.environ.setdefault(key, value)
+
+
 # Load environment variables from .env file
-load_dotenv()
+_load_dotenv_skipping_blanks()
 
 # Auto-configure cache from settings if enabled
 from .infra.settings import settings as _settings

{prompture-1.11.2.dev3 → prompture-1.12.1.dev1}/prompture/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '1.11.2.dev3'
-__version_tuple__ = version_tuple = (1, 11, 2, 'dev3')
+__version__ = version = '1.12.1.dev1'
+__version_tuple__ = version_tuple = (1, 12, 1, 'dev1')
 
 __commit_id__ = commit_id = None

{prompture-1.11.2.dev3 → prompture-1.12.1.dev1}/prompture/drivers/__init__.py

@@ -47,6 +47,8 @@ from .async_groq_driver import AsyncGroqDriver
 from .async_hugging_driver import AsyncHuggingFaceDriver
 from .async_img_gen_base import AsyncImageGenDriver
 from .async_jina_rerank_driver import AsyncJinaRerankDriver
+from .async_kev_decision_driver import AsyncKevDecisionDriver
+from .async_laya_decision_driver import AsyncLayaDecisionDriver
 from .async_lmstudio_driver import AsyncLMStudioDriver
 from .async_local_http_driver import AsyncLocalHTTPDriver
 from .async_mistral_moderation_driver import AsyncMistralModerationDriver
@@ -66,7 +68,9 @@ from .async_runway_tts_driver import AsyncRunwayTTSDriver
 from .async_runway_video_gen_driver import AsyncRunwayVideoGenDriver
 from .async_stability_img_gen_driver import AsyncStabilityImageGenDriver
 from .async_stt_base import AsyncSTTDriver
+from .async_systemone_compatible_driver import AsyncSystemOneCompatibleDriver
 from .async_tts_base import AsyncTTSDriver
+from .async_typesafe_decision_driver import AsyncTypeSafeDecisionDriver
 from .async_video_gen_base import AsyncVideoGenDriver
 from .async_voyage_rerank_driver import AsyncVoyageRerankDriver
 from .async_zai_driver import AsyncZaiDriver
@@ -80,6 +84,28 @@ from .azure_driver import AzureDriver
 from .cachibot_driver import CachiBotDriver
 from .claude_driver import ClaudeDriver
 from .cohere_rerank_driver import CohereRerankDriver
+from .decision_base import (
+    QUESTION_TYPES,
+    Answer,
+    AsyncDecisionDriver,
+    Choice,
+    ChoiceAnswer,
+    DecisionDriver,
+    DecisionResponse,
+    Noul,
+    NoulAnswer,
+    Question,
+    Score,
+    ScoreAnswer,
+    calculate_decision_cost,
+    normalize_questions,
+)
+from .decision_registry import (
+    ASYNC_DECISION_DRIVER_REGISTRY,
+    DECISION_DRIVER_REGISTRY,
+    get_async_decision_driver_for_model,
+    get_decision_driver_for_model,
+)
 from .elevenlabs_stt_driver import ElevenLabsSTTDriver
 from .elevenlabs_tts_driver import ElevenLabsTTSDriver
 from .embedding_base import EMBEDDING_MODEL_DIMENSIONS, EmbeddingDriver
@@ -92,6 +118,8 @@ from .groq_driver import GroqDriver
 from .hugging_driver import HuggingFaceDriver
 from .img_gen_base import ImageGenDriver
 from .jina_rerank_driver import JinaRerankDriver
+from .kev_decision_driver import KevDecisionDriver
+from .laya_decision_driver import LayaDecisionDriver
 from .lmstudio_driver import LMStudioDriver
 from .local_http_driver import LocalHTTPDriver
 from .mistral_moderation_driver import MistralModerationDriver
@@ -121,6 +149,7 @@ from .provider_descriptors import (
 )
 from .registry import (
     _get_sync_registry,
+    get_async_decision_driver_factory,
     get_async_driver_factory,
     get_async_embedding_driver_factory,
     get_async_img_gen_driver_factory,
@@ -129,6 +158,7 @@ from .registry import (
     get_async_stt_driver_factory,
     get_async_tts_driver_factory,
     get_async_video_gen_driver_factory,
+    get_decision_driver_factory,
     get_driver_factory,
     get_embedding_driver_factory,
     get_img_gen_driver_factory,
@@ -137,6 +167,7 @@ from .registry import (
     get_stt_driver_factory,
     get_tts_driver_factory,
     get_video_gen_driver_factory,
+    is_async_decision_driver_registered,
     is_async_driver_registered,
     is_async_embedding_driver_registered,
     is_async_img_gen_driver_registered,
@@ -145,6 +176,7 @@ from .registry import (
     is_async_stt_driver_registered,
     is_async_tts_driver_registered,
     is_async_video_gen_driver_registered,
+    is_decision_driver_registered,
     is_driver_registered,
     is_embedding_driver_registered,
     is_img_gen_driver_registered,
@@ -153,6 +185,7 @@ from .registry import (
     is_stt_driver_registered,
     is_tts_driver_registered,
     is_video_gen_driver_registered,
+    list_registered_async_decision_drivers,
     list_registered_async_drivers,
     list_registered_async_embedding_drivers,
     list_registered_async_img_gen_drivers,
@@ -161,6 +194,7 @@ from .registry import (
     list_registered_async_stt_drivers,
     list_registered_async_tts_drivers,
     list_registered_async_video_gen_drivers,
+    list_registered_decision_drivers,
     list_registered_drivers,
     list_registered_embedding_drivers,
     list_registered_img_gen_drivers,
@@ -170,6 +204,7 @@ from .registry import (
     list_registered_tts_drivers,
     list_registered_video_gen_drivers,
     load_entry_point_drivers,
+    register_async_decision_driver,
     register_async_driver,
     register_async_embedding_driver,
     register_async_img_gen_driver,
@@ -178,6 +213,7 @@ from .registry import (
     register_async_stt_driver,
     register_async_tts_driver,
     register_async_video_gen_driver,
+    register_decision_driver,
     register_driver,
     register_embedding_driver,
     register_img_gen_driver,
@@ -186,6 +222,7 @@ from .registry import (
     register_stt_driver,
     register_tts_driver,
     register_video_gen_driver,
+    unregister_async_decision_driver,
     unregister_async_driver,
     unregister_async_embedding_driver,
     unregister_async_img_gen_driver,
@@ -194,6 +231,7 @@ from .registry import (
     unregister_async_stt_driver,
     unregister_async_tts_driver,
     unregister_async_video_gen_driver,
+    unregister_decision_driver,
     unregister_driver,
     unregister_embedding_driver,
     unregister_img_gen_driver,
@@ -222,7 +260,9 @@ from .runway_tts_driver import RunwayTTSDriver
 from .runway_video_gen_driver import RunwayVideoGenDriver
 from .stability_img_gen_driver import StabilityImageGenDriver
 from .stt_base import STTDriver
+from .systemone_compatible_driver import SystemOneCompatibleDriver
 from .tts_base import TTSDriver
+from .typesafe_decision_driver import TypeSafeDecisionDriver
 from .video_gen_base import VideoGenDriver
 from .voyage_rerank_driver import VoyageRerankDriver
 from .zai_driver import ZaiDriver
@@ -511,11 +551,13 @@ def provider_for_model(model_str: str, *, canonical: bool = False) -> str:
 
 
 __all__ = [
+    "ASYNC_DECISION_DRIVER_REGISTRY",
     "ASYNC_DRIVER_REGISTRY",
     "ASYNC_MODERATION_DRIVER_REGISTRY",
     # Provider driver maps (for explicit-credential construction)
     "ASYNC_PROVIDER_DRIVER_MAP",
     "ASYNC_RERANK_DRIVER_REGISTRY",
+    "DECISION_DRIVER_REGISTRY",
     # Legacy registry dicts (for backwards compatibility)
     "DRIVER_REGISTRY",
     # Embedding model dimension metadata
@@ -529,12 +571,15 @@ __all__ = [
     "PROVIDER_DRIVER_MAP",
     # Provider name mapping
     "PROVIDER_NAME_MAP",
+    "QUESTION_TYPES",
     "RERANK_DRIVER_REGISTRY",
     "RUNWAY_ALL_MODALITIES",
     "RUNWAY_ALL_OPERATIONS",
     "RUNWAY_MODEL_INFO",
     # Sync LLM drivers
     "AirLLMDriver",
+    # Decision base + question/answer types
+    "Answer",
     # Async LLM drivers
     "AsyncAirLLMDriver",
     "AsyncAzureDriver",
@@ -542,6 +587,7 @@ __all__ = [
     "AsyncClaudeDriver",
     # Async rerank drivers
     "AsyncCohereRerankDriver",
+    "AsyncDecisionDriver",
     # Async base classes
     "AsyncDriver",
     # Async audio drivers
@@ -559,7 +605,9 @@ __all__ = [
     "AsyncHuggingFaceDriver",
     "AsyncImageGenDriver",
     "AsyncJinaRerankDriver",
+    "AsyncKevDecisionDriver",
     "AsyncLMStudioDriver",
+    "AsyncLayaDecisionDriver",
     "AsyncLocalHTTPDriver",
     "AsyncMistralModerationDriver",
     "AsyncModelScopeDriver",
@@ -580,16 +628,22 @@ __all__ = [
     "AsyncRunwayVideoGenDriver",
     "AsyncSTTDriver",
     "AsyncStabilityImageGenDriver",
+    "AsyncSystemOneCompatibleDriver",
     "AsyncTTSDriver",
+    "AsyncTypeSafeDecisionDriver",
     "AsyncVideoGenDriver",
     "AsyncVoyageRerankDriver",
     "AsyncZaiDriver",
     # Sync LLM drivers
     "AzureDriver",
     "CachiBotDriver",
+    "Choice",
+    "ChoiceAnswer",
     "ClaudeDriver",
     # Sync rerank drivers
     "CohereRerankDriver",
+    "DecisionDriver",
+    "DecisionResponse",
     # Sync audio drivers
     "ElevenLabsSTTDriver",
     "ElevenLabsTTSDriver",
@@ -606,7 +660,9 @@ __all__ = [
     # Image gen base class
     "ImageGenDriver",
     "JinaRerankDriver",
+    "KevDecisionDriver",
     "LMStudioDriver",
+    "LayaDecisionDriver",
     "LocalHTTPDriver",
     "MistralModerationDriver",
     "ModelScopeDriver",
@@ -614,6 +670,8 @@ __all__ = [
     "ModerationDriver",
     "ModerationResult",
     "MoonshotDriver",
+    "Noul",
+    "NoulAnswer",
     "OllamaDriver",
     "OllamaEmbeddingDriver",
     "OpenAIDriver",
@@ -625,6 +683,7 @@ __all__ = [
     "OpenRouterDriver",
     # Branding (display metadata)
     "ProviderBrand",
+    "Question",
     # Rerank base + result types
     "RerankDriver",
     "RerankResult",
@@ -634,13 +693,20 @@ __all__ = [
     "RunwayVideoGenDriver",
     # STT/TTS base classes
     "STTDriver",
+    "Score",
+    "ScoreAnswer",
     "StabilityImageGenDriver",
+    "SystemOneCompatibleDriver",
     "TTSDriver",
+    "TypeSafeDecisionDriver",
     "VideoGenDriver",
     "VoyageRerankDriver",
     "ZaiDriver",
+    "calculate_decision_cost",
     # Azure config API
     "clear_azure_configs",
+    "get_async_decision_driver_factory",
+    "get_async_decision_driver_for_model",
     "get_async_driver",
     "get_async_driver_for_model",
     # Embedding registry query functions
@@ -672,6 +738,8 @@ __all__ = [
     # Video gen factory functions
     "get_async_video_gen_driver_for_model",
     # LLM factory functions
+    "get_decision_driver_factory",
+    "get_decision_driver_for_model",
     "get_driver",
     "get_driver_for_model",
     "get_embedding_driver_factory",
@@ -696,6 +764,7 @@ __all__ = [
     "get_video_gen_driver_for_model",
     "icon_url",
     # Other registry query functions
+    "is_async_decision_driver_registered",
     "is_async_driver_registered",
     "is_async_embedding_driver_registered",
     "is_async_img_gen_driver_registered",
@@ -704,6 +773,7 @@ __all__ = [
     "is_async_stt_driver_registered",
     "is_async_tts_driver_registered",
     "is_async_video_gen_driver_registered",
+    "is_decision_driver_registered",
     "is_driver_registered",
     "is_embedding_driver_registered",
     "is_img_gen_driver_registered",
@@ -712,6 +782,7 @@ __all__ = [
     "is_stt_driver_registered",
     "is_tts_driver_registered",
     "is_video_gen_driver_registered",
+    "list_registered_async_decision_drivers",
     "list_registered_async_drivers",
     "list_registered_async_embedding_drivers",
     "list_registered_async_img_gen_drivers",
@@ -720,6 +791,7 @@ __all__ = [
     "list_registered_async_stt_drivers",
     "list_registered_async_tts_drivers",
     "list_registered_async_video_gen_drivers",
+    "list_registered_decision_drivers",
     "list_registered_drivers",
     "list_registered_embedding_drivers",
     "list_registered_img_gen_drivers",
@@ -729,9 +801,11 @@ __all__ = [
     "list_registered_tts_drivers",
     "list_registered_video_gen_drivers",
     "load_entry_point_drivers",
+    "normalize_questions",
     "parse_model_string",
     "provider_for_model",
     "register_all_builtin_drivers",
+    "register_async_decision_driver",
     "register_async_driver",
     "register_async_embedding_driver",
     "register_async_img_gen_driver",
@@ -742,6 +816,7 @@ __all__ = [
     "register_async_video_gen_driver",
     "register_azure_config",
     # Registry functions (public API)
+    "register_decision_driver",
     "register_driver",
     "register_embedding_driver",
     "register_img_gen_driver",
@@ -751,6 +826,7 @@ __all__ = [
     "register_tts_driver",
     "register_video_gen_driver",
     "set_azure_config_resolver",
+    "unregister_async_decision_driver",
     "unregister_async_driver",
     "unregister_async_embedding_driver",
     "unregister_async_img_gen_driver",
@@ -760,6 +836,7 @@ __all__ = [
     "unregister_async_tts_driver",
     "unregister_async_video_gen_driver",
     "unregister_azure_config",
+    "unregister_decision_driver",
     "unregister_driver",
     "unregister_embedding_driver",
     "unregister_img_gen_driver",

prompture-1.12.1.dev1/prompture/drivers/async_kev_decision_driver.py

@@ -0,0 +1,19 @@
+"""Async Kev decision driver — self-hosted Jev-compatible decision models."""
+
+from __future__ import annotations
+
+from .async_systemone_compatible_driver import AsyncSystemOneCompatibleDriver
+from .kev_decision_driver import KevDecisionDriver
+
+
+class AsyncKevDecisionDriver(AsyncSystemOneCompatibleDriver):
+    """Async self-hosted Kev decision driver."""
+
+    PROVIDER = KevDecisionDriver.PROVIDER
+    DEFAULT_BASE_URL = KevDecisionDriver.DEFAULT_BASE_URL
+    DEFAULT_MODEL = KevDecisionDriver.DEFAULT_MODEL
+    ENV_KEY = KevDecisionDriver.ENV_KEY
+    ENV_BASE_URL = KevDecisionDriver.ENV_BASE_URL
+    REQUIRES_API_KEY = False
+
+    KNOWN_MODELS: tuple[str, ...] = KevDecisionDriver.KNOWN_MODELS