agent-framework-lib 0.8.15__tar.gz → 0.8.17__tar.gz

{agent_framework_lib-0.8.15/agent_framework_lib.egg-info → agent_framework_lib-0.8.17}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agent-framework-lib
-Version: 0.8.15
+Version: 0.8.17
 Summary: A comprehensive Python framework for building and serving conversational AI agents with FastAPI
 Author-email: Sebastian Pavel <sebastian@cinco.ai>, Elliott Girard <elliott.girard@icloud.com>
 Maintainer-email: Sebastian Pavel <sebastian@cinco.ai>
@@ -333,7 +333,7 @@ Pour **déployer** un agent, partez de l'image officielle — Chromium Playwrigh
 sandbox locale déjà en place (sinon l'agent ne peut ni rendre un PNG ni installer quoi que ce soit) :
 
 ```dockerfile
-FROM ghcr.io/komus-ai/agent-framework-runtime:0.8.15
+FROM ghcr.io/komus-ai/agent-framework-runtime:0.8.17
 ```
 
 Détails, publication et visibilité GHCR : [docs/AGENT_RUNTIME_IMAGE.md](docs/AGENT_RUNTIME_IMAGE.md).
@@ -542,9 +542,9 @@ DEFAULT_MODEL_MODE=auto
 AUTO_CLASSIFIER_MODEL=gpt-5.4-nano
 
 # Preferred models per tier (comma-separated, in order of preference)
-PREFERRED_LIGHT_MODELS=gpt-5.4-nano,claude-haiku-4-5-20251001,gemini-2.5-flash-lite
-PREFERRED_STANDARD_MODELS=gpt-5.4-mini,gemini-2.5-flash
-PREFERRED_ADVANCED_MODELS=gpt-5.2,claude-opus-4-6,gemini-2.5-pro
+PREFERRED_LIGHT_MODELS=gpt-5.4-nano,claude-haiku-4-5-20251001,gemini-3.5-flash-lite
+PREFERRED_STANDARD_MODELS=gpt-5.4-mini,gemini-3.6-flash
+PREFERRED_ADVANCED_MODELS=gpt-5.2,claude-opus-4-6,gemini-3.1-pro-preview
 ```
 
 ### API Endpoint
@@ -596,7 +596,7 @@ class MyAgent(LlamaIndexAgent):
 **Runtime Configuration:**
 
 Users can override settings per session via the API or web UI:
-- Model selection (gpt-5.2, claude-sonnet-4-6, gemini-2.5-pro)
+- Model selection (gpt-5.2, claude-sonnet-4-6, gemini-3.1-pro-preview)
 - Temperature (0.0 - 1.0)
 - Max tokens
 - System prompt override

{agent_framework_lib-0.8.15 → agent_framework_lib-0.8.17}/README.md

@@ -173,7 +173,7 @@ Pour **déployer** un agent, partez de l'image officielle — Chromium Playwrigh
 sandbox locale déjà en place (sinon l'agent ne peut ni rendre un PNG ni installer quoi que ce soit) :
 
 ```dockerfile
-FROM ghcr.io/komus-ai/agent-framework-runtime:0.8.15
+FROM ghcr.io/komus-ai/agent-framework-runtime:0.8.17
 ```
 
 Détails, publication et visibilité GHCR : [docs/AGENT_RUNTIME_IMAGE.md](docs/AGENT_RUNTIME_IMAGE.md).
@@ -382,9 +382,9 @@ DEFAULT_MODEL_MODE=auto
 AUTO_CLASSIFIER_MODEL=gpt-5.4-nano
 
 # Preferred models per tier (comma-separated, in order of preference)
-PREFERRED_LIGHT_MODELS=gpt-5.4-nano,claude-haiku-4-5-20251001,gemini-2.5-flash-lite
-PREFERRED_STANDARD_MODELS=gpt-5.4-mini,gemini-2.5-flash
-PREFERRED_ADVANCED_MODELS=gpt-5.2,claude-opus-4-6,gemini-2.5-pro
+PREFERRED_LIGHT_MODELS=gpt-5.4-nano,claude-haiku-4-5-20251001,gemini-3.5-flash-lite
+PREFERRED_STANDARD_MODELS=gpt-5.4-mini,gemini-3.6-flash
+PREFERRED_ADVANCED_MODELS=gpt-5.2,claude-opus-4-6,gemini-3.1-pro-preview
 ```
 
 ### API Endpoint
@@ -436,7 +436,7 @@ class MyAgent(LlamaIndexAgent):
 **Runtime Configuration:**
 
 Users can override settings per session via the API or web UI:
-- Model selection (gpt-5.2, claude-sonnet-4-6, gemini-2.5-pro)
+- Model selection (gpt-5.2, claude-sonnet-4-6, gemini-3.1-pro-preview)
 - Temperature (0.0 - 1.0)
 - Max tokens
 - System prompt override

{agent_framework_lib-0.8.15 → agent_framework_lib-0.8.17}/agent_framework/core/context_budget.py

@@ -34,12 +34,17 @@ class ModelContextRegistry:
         # Light tier
         "gpt-5.4-nano": 400_000,
         "gemini-2.5-flash-lite": 1_048_576,
-        "gemini-3.1-flash-lite-preview": 1_048_576,
+        "gemini-3.5-flash-lite": 1_048_576,
+        "gemini-3.1-flash-lite": 1_048_576,
         # Standard tier
         "gpt-5.4-mini": 400_000,
         "claude-haiku-4-5-20251001": 200_000,
         "gemini-2.5-flash": 1_048_576,
         "gemini-3-flash-preview": 1_048_576,
+        "gemini-3.5-flash": 1_048_576,
+        "gemini-3.6-flash": 1_048_576,
+        "gemini-3.7-flash": 1_048_576,
+        "gemini-3.8-flash": 1_048_576,
         # Advanced tier
         "gpt-5.2": 272_000,
         "gpt-5.4": 1_050_000,
@@ -58,6 +63,13 @@ class ModelContextRegistry:
         "claude-fable-5-1": 1_000_000,
         "gpt-5.4-pro": 1_050_000,
         "gpt-5.5-pro": 400_000,
+        "gpt-5.6-sol": 1_050_000,
+        "gpt-5.6-terra": 1_050_000,
+        "gpt-5.6-luna": 1_050_000,
+        # Famille GPT-6 (doc OpenAI 29/09/2026) : 1,05 M de contexte.
+        "gpt-6-luna": 1_050_000,
+        "gpt-6-sol": 1_050_000,
+        "gpt-6-astra": 1_050_000,
         "gemini-2.5-pro": 1_048_576,
         "gemini-3.1-pro-preview": 1_048_576,
         # Mistral — valeurs max_context_length de GET /v1/models, compte Komus, 17/09/2026.

{agent_framework_lib-0.8.15 → agent_framework_lib-0.8.17}/agent_framework/core/model_clients.py

@@ -20,6 +20,7 @@ from .model_config import (
     ModelProvider,
     anthropic_supports_sampling,
     model_config,
+    openai_uses_responses_api,
 )
 
 
@@ -627,6 +628,36 @@ def register_anthropic_model_with_llamaindex(model_name: str | None) -> bool:
     return True
 
 
+def register_openai_model_with_llamaindex(model_name: str | None) -> bool:
+    """Ajoute ``model_name`` aux tables de ``llama_index.llms.openai.utils`` si elles l'ignorent.
+
+    Même défaut que côté Anthropic : llama-index lève ``Unknown model`` pour tout id absent
+    de ses tables figées (``gpt-6-*`` le 29/09/2026). Le modèle est enregistré comme modèle
+    de raisonnement (``O1_MODELS`` : ``max_completion_tokens``, ``reasoning_effort``), ce que
+    sont tous les GPT-5+. Fenêtre de contexte : ``ModelContextRegistry`` (repli 400k).
+    """
+    if not model_name:
+        return False
+    try:
+        from llama_index.llms.openai import utils as li_utils
+    except ImportError:
+        return False
+    if model_name in li_utils.ALL_AVAILABLE_MODELS:
+        return False
+    from .context_budget import ModelContextRegistry
+
+    context_window = ModelContextRegistry.CONTEXT_WINDOWS.get(model_name) or 400_000
+    for table in (li_utils.O1_MODELS, li_utils.ALL_AVAILABLE_MODELS, li_utils.CHAT_MODELS):
+        table[model_name] = context_window
+    logger.info(
+        "[ModelClientFactory] modèle %s inconnu de llama-index-llms-openai : enregistré "
+        "(fenêtre %d)",
+        model_name,
+        context_window,
+    )
+    return True
+
+
 _SAMPLING_DROP_WARNED: set[tuple[str, str]] = set()
 
 
@@ -1547,6 +1578,10 @@ class ModelClientFactory:
                 f"Install with: uv add llama-index-llms-openai"
             )
 
+        register_openai_model_with_llamaindex(params["model"])
+        if openai_uses_responses_api(params["model"]):
+            return self._create_llamaindex_openai_responses(params)
+
         logger.info(
             f"[ModelClientFactory] Creating LlamaIndex OpenAI LLM for model: {params['model']}"
         )
@@ -1566,6 +1601,35 @@ class ModelClientFactory:
 
         return llm
 
+    def _create_llamaindex_openai_responses(self, params: dict[str, Any]) -> Any:
+        """LLM LlamaIndex sur l'API Responses (voir ``OPENAI_RESPONSES_API_MODELS``).
+
+        ``max_tokens`` s'y appelle ``max_output_tokens`` ; la fenêtre de contexte est
+        passée explicitement. Pas de patch de sanitisation : il vise les ``messages`` de
+        /v1/chat/completions, llama-index sérialise lui-même les ``input`` de Responses.
+        """
+        from llama_index.llms.openai import OpenAIResponses
+
+        from .context_budget import ModelContextRegistry
+
+        params = dict(params)
+        if "max_tokens" in params:
+            params["max_output_tokens"] = params.pop("max_tokens")
+        params.setdefault(
+            "context_window", ModelContextRegistry.CONTEXT_WINDOWS.get(params["model"])
+        )
+        logger.info(
+            "[ModelClientFactory] Creating LlamaIndex OpenAI Responses LLM for model: %s",
+            params["model"],
+        )
+        try:
+            return OpenAIResponses(**params)
+        except TypeError as e:
+            logger.warning(
+                f"[ModelClientFactory] Parameter error: {e}, retrying without problematic params"
+            )
+            return self._retry_without_unsupported_params(OpenAIResponses, params, e)
+
     def _create_llamaindex_anthropic(self, params: dict[str, Any]) -> Any:
         """
         Create a LlamaIndex Anthropic LLM with error handling.

{agent_framework_lib-0.8.15 → agent_framework_lib-0.8.17}/agent_framework/core/model_config.py

@@ -114,8 +114,10 @@ class ModelConfigManager:
     """
 
     # Default model mappings (can be overridden by environment variables)
-    # Modèles constatés sur les comptes Komus le 15/09/2026 (GET /v1/models des deux
-    # fournisseurs). Un id absent d'ici reste routé par motif (« gpt », « claude »).
+    # Modèles constatés sur les comptes Komus le 15/09/2026, revus le 29/09/2026 : seuls
+    # les ids appelables à travers la lib (chat + stream + outil) sont proposés. Un id absent
+    # d'ici reste routé par motif (« gpt », « claude », « gemini »). Les modèles de
+    # ``OPENAI_RESPONSES_API_MODELS`` passent par l'API Responses (voir plus bas).
     DEFAULT_OPENAI_MODELS: Final[list[str]] = [
         "gpt-5.4-nano",
         "gpt-5.4-mini",
@@ -126,6 +128,9 @@ class ModelConfigManager:
         "gpt-5.6-sol",
         "gpt-5.6-terra",
         "gpt-5.6-luna",
+        "gpt-6-luna",
+        "gpt-6-sol",
+        "gpt-6-astra",
     ]
 
     DEFAULT_ANTHROPIC_MODELS: Final[list[str]] = [
@@ -143,12 +148,17 @@ class ModelConfigManager:
         "claude-fable-5-1",
     ]
 
+    # Gemini sondés le 29/09/2026 à travers la lib (chat + stream + outil, tous OK).
+    # gemini-3.1-flash-lite-preview est arrêté ; la famille 2.5 répond 404 « no longer
+    # available to new users » sur le compte Komus.
     DEFAULT_GEMINI_MODELS: Final[list[str]] = [
-        "gemini-2.5-flash-lite",
-        "gemini-3.1-flash-lite-preview",
-        "gemini-2.5-flash",
+        "gemini-3.1-flash-lite",
+        "gemini-3.5-flash-lite",
+        "gemini-3.5-flash",
+        "gemini-3.6-flash",
+        "gemini-3.7-flash",
+        "gemini-3.8-flash",
         "gemini-3-flash-preview",
-        "gemini-2.5-pro",
         "gemini-3.1-pro-preview",
     ]
 
@@ -1032,6 +1042,26 @@ ANTHROPIC_NO_SAMPLING_MODELS: Final[tuple[str, ...]] = (
 ANTHROPIC_SAMPLING_PARAMS: Final[tuple[str, ...]] = ("temperature", "top_p", "top_k")
 
 
+# ---------------------------------------------------------------------------
+# OpenAI : modèles servis par l'API Responses
+# ---------------------------------------------------------------------------
+# Sur /v1/chat/completions, les « pro » n'existent pas et GPT-5.6 / GPT-6 refusent les
+# outils dès qu'ils raisonnent (400 « Function tools with reasoning_effort are not
+# supported », constaté le 29/09/2026). L'API Responses accepte outils + raisonnement :
+# ces modèles y passent (``OpenAIResponses`` de llama-index). Correspondance par préfixe.
+OPENAI_RESPONSES_API_MODELS: Final[tuple[str, ...]] = (
+    "gpt-6",
+    "gpt-5.6",
+    "gpt-5.5-pro",
+    "gpt-5.4-pro",
+)
+
+
+def openai_uses_responses_api(model_name: str | None) -> bool:
+    """True si le modèle OpenAI doit passer par l'API Responses."""
+    return bool(model_name) and model_name.lower().startswith(OPENAI_RESPONSES_API_MODELS)
+
+
 def anthropic_supports_sampling(model_name: str | None) -> bool:
     """True si le modèle Anthropic accepte encore temperature / top_p / top_k.
 

{agent_framework_lib-0.8.15 → agent_framework_lib-0.8.17}/agent_framework/core/model_router.py

@@ -132,11 +132,13 @@ class ModelRouter:
     # les autres modèles restent en repli si la clé Anthropic est absente ou si
     # aucun modèle Claude du tier n'est disponible.
     # Mistral en repli de fin de liste (intégré le 17/09/2026) : jamais devant Claude.
+    # 29/09/2026 : OpenAI vérifié à travers la lib (chat + stream + outil) ; Gemini sondé de même
+    # (clé de l'env dev Mozart) ; gpt-5.5-pro sorti (Responses API seulement).
     DEFAULT_LIGHT_MODELS: list[str] = [
         "claude-haiku-4-5-20251001",
         "gpt-5.4-nano",
-        "gemini-2.5-flash-lite",
-        "gemini-3.1-flash-lite-preview",
+        "gemini-3.5-flash-lite",
+        "gemini-3.1-flash-lite",
         "ministral-8b-latest",
     ]
     DEFAULT_STANDARD_MODELS: list[str] = [
@@ -144,8 +146,8 @@ class ModelRouter:
         "claude-sonnet-5",
         "claude-sonnet-4-6",
         "gpt-5.4-mini",
-        "gemini-2.5-flash",
-        "gemini-3-flash-preview",
+        "gemini-3.6-flash",
+        "gemini-3.5-flash",
         "ministral-14b-latest",
     ]
     DEFAULT_ADVANCED_MODELS: list[str] = [
@@ -154,14 +156,14 @@ class ModelRouter:
         "claude-fable-5-1",
         "claude-opus-4-8",
         "gpt-5.5",
-        "gpt-5.5-pro",
         "claude-sonnet-5-5",
         "claude-sonnet-5",
         "claude-sonnet-4-6",
         "claude-opus-4-6",
         "claude-opus-4-5-20251101",
         "gpt-5.4",
-        "gemini-2.5-pro",
+        "gemini-3.8-flash",
+        "gemini-3.7-flash",
         "gemini-3.1-pro-preview",
         # Un modèle est jugé disponible dès que la clé du provider existe — le routeur ne
         # sait pas si le PLAN du compte le sert (429 quota 0 / 403 sinon, sans repli). Seuls

{agent_framework_lib-0.8.15 → agent_framework_lib-0.8.17}/agent_framework/monitoring/token_counter.py

@@ -115,7 +115,12 @@ class TokenCounter:
         # Google Gemini models (cl100k_base approximation)
         "gemini-3.1-pro-preview": "cl100k_base",
         "gemini-3-flash-preview": "cl100k_base",
-        "gemini-3.1-flash-lite-preview": "cl100k_base",
+        "gemini-3.1-flash-lite": "cl100k_base",
+        "gemini-3.5-flash-lite": "cl100k_base",
+        "gemini-3.5-flash": "cl100k_base",
+        "gemini-3.6-flash": "cl100k_base",
+        "gemini-3.7-flash": "cl100k_base",
+        "gemini-3.8-flash": "cl100k_base",
         "gemini-2.5-pro": "cl100k_base",
         "gemini-2.5-flash": "cl100k_base",
         "gemini-2.5-flash-lite": "cl100k_base",
@@ -140,6 +145,7 @@ class TokenCounter:
 
     # Pattern-based encoding detection for unknown models
     PROVIDER_ENCODINGS: dict[str, str] = {
+        "gpt-6": "o200k_base",  # GPT-6 series (o200k supposé, comme GPT-5)
         "gpt-5": "o200k_base",  # GPT-5 series
         "gpt-4o": "o200k_base",  # GPT-4o series
         "gpt-4": "cl100k_base",  # GPT-4 series

{agent_framework_lib-0.8.15 → agent_framework_lib-0.8.17}/agent_framework/web/docs/Dockerfile

@@ -2,7 +2,7 @@
 # Base = image de runtime officielle (Chromium Playwright, Deno, git, root, sandbox locale) :
 # voir docs/AGENT_RUNTIME_IMAGE.md. Ne repartez PAS de python:*-slim à la main : vous
 # retomberiez sur une sandbox sans libs Chromium ni droits (export PNG impossible).
-FROM ghcr.io/komus-ai/agent-framework-runtime:0.8.15
+FROM ghcr.io/komus-ai/agent-framework-runtime:0.8.17
 
 WORKDIR /app
 

{agent_framework_lib-0.8.15 → agent_framework_lib-0.8.17}/agent_framework/web/docs/README.md

@@ -173,7 +173,7 @@ Pour **déployer** un agent, partez de l'image officielle — Chromium Playwrigh
 sandbox locale déjà en place (sinon l'agent ne peut ni rendre un PNG ni installer quoi que ce soit) :
 
 ```dockerfile
-FROM ghcr.io/komus-ai/agent-framework-runtime:0.8.15
+FROM ghcr.io/komus-ai/agent-framework-runtime:0.8.17
 ```
 
 Détails, publication et visibilité GHCR : [docs/AGENT_RUNTIME_IMAGE.md](docs/AGENT_RUNTIME_IMAGE.md).
@@ -382,9 +382,9 @@ DEFAULT_MODEL_MODE=auto
 AUTO_CLASSIFIER_MODEL=gpt-5.4-nano
 
 # Preferred models per tier (comma-separated, in order of preference)
-PREFERRED_LIGHT_MODELS=gpt-5.4-nano,claude-haiku-4-5-20251001,gemini-2.5-flash-lite
-PREFERRED_STANDARD_MODELS=gpt-5.4-mini,gemini-2.5-flash
-PREFERRED_ADVANCED_MODELS=gpt-5.2,claude-opus-4-6,gemini-2.5-pro
+PREFERRED_LIGHT_MODELS=gpt-5.4-nano,claude-haiku-4-5-20251001,gemini-3.5-flash-lite
+PREFERRED_STANDARD_MODELS=gpt-5.4-mini,gemini-3.6-flash
+PREFERRED_ADVANCED_MODELS=gpt-5.2,claude-opus-4-6,gemini-3.1-pro-preview
 ```
 
 ### API Endpoint
@@ -436,7 +436,7 @@ class MyAgent(LlamaIndexAgent):
 **Runtime Configuration:**
 
 Users can override settings per session via the API or web UI:
-- Model selection (gpt-5.2, claude-sonnet-4-6, gemini-2.5-pro)
+- Model selection (gpt-5.2, claude-sonnet-4-6, gemini-3.1-pro-preview)
 - Temperature (0.0 - 1.0)
 - Max tokens
 - System prompt override

{agent_framework_lib-0.8.15 → agent_framework_lib-0.8.17/agent_framework/web}/docs/configuration.md

@@ -108,47 +108,30 @@ MISTRAL_API_KEY=your-mistral-key
 - **Default**: `claude-haiku-4-5-20251001`
 - **Description**: Default model to use when no model is specified in agent configuration. The framework automatically detects the provider from the model name.
 
-**Supported Models:**
+**Supported Models** (checked 29/09/2026 through the lib: chat, streaming and tool call):
+
+**Anthropic:**
+- `claude-opus-5-5`, `claude-opus-5`, `claude-opus-4-8`, `claude-opus-4-7`, `claude-opus-4-6`, `claude-opus-4-5-20251101`
+- `claude-sonnet-5-5`, `claude-sonnet-5`, `claude-sonnet-4-6`
+- `claude-haiku-4-5-20251001`
+- `claude-fable-5-1`, `claude-fable-5`
 
 **OpenAI:**
-- `gpt-5.2` - GPT-5.2 (latest flagship)
-- `gpt-5` - GPT-5
-- `gpt-5.4-mini` - Smaller, fast
-- `gpt-5.4-nano` - Ultra-lightweight, cost-efficient
-- `o1` - O1
-
-**Anthropic (Claude 4 Family - Latest):**
-- `claude-opus-4-6` - Claude Opus 4.6 (most powerful, adaptive thinking, 1M context beta, released February 2026)
-- `claude-opus-4-5-20251101` - Claude Opus 4.5
-- `claude-opus-4-1-20250514` - Claude Opus 4.1 (incremental update to Opus 4)
-- `claude-sonnet-4-6` - Claude Sonnet 4.6 (latest Sonnet, balanced performance)
-- `claude-haiku-4-5-20251001` - Claude Haiku 4.5 (fastest, most cost-efficient)
-
-**Anthropic (Claude 3 Family - Previous Generation):**
-- `claude-3-7-sonnet-20250219` - Claude 3.7 Sonnet
-- `claude-3-5-sonnet-20241022` - Claude 3.5 Sonnet
-- `claude-3-5-haiku-20241022` - Claude 3.5 Haiku
-
-**Google Gemini (Gemini 3 Family - Latest):**
-- `gemini-3-pro-preview` - Gemini 3 Pro (most intelligent, state-of-the-art reasoning)
-- `gemini-3-deep-think` - Gemini 3 Deep Think (advanced reasoning mode)
-
-**Google Gemini (Gemini 2.5 Family):**
-- `gemini-2.5-pro` - Gemini 2.5 Pro (powerful with adaptive thinking)
-- `gemini-2.5-flash` - Gemini 2.5 Flash (fast, well-rounded capabilities)
-- `gemini-2.5-flash-lite` - Gemini 2.5 Flash-Lite (optimized for cost-efficiency)
-
-**Google Gemini (Gemini 2.0 Family):**
-- `gemini-2.0-flash` - Gemini 2.0 Flash (1M token context window)
-- `gemini-2.0-flash-thinking` - Gemini 2.0 Flash Thinking (shows reasoning process)
+- `gpt-6-sol`, `gpt-6-luna`, `gpt-6-astra`, `gpt-5.6-sol`, `gpt-5.6-terra`, `gpt-5.6-luna`, `gpt-5.5-pro`, `gpt-5.4-pro` (served through the OpenAI Responses API)
+- `gpt-5.5`, `gpt-5.4`, `gpt-5.4-mini`, `gpt-5.4-nano`
+
+**Google Gemini:**
+- `gemini-3.8-flash`, `gemini-3.7-flash`, `gemini-3.6-flash`, `gemini-3.5-flash`, `gemini-3-flash-preview`
+- `gemini-3.5-flash-lite`, `gemini-3.1-flash-lite`
+- `gemini-3.1-pro-preview`
 
 **Example:**
 ```bash
 DEFAULT_MODEL=claude-haiku-4-5-20251001
 # or
-DEFAULT_MODEL=claude-sonnet-4-6
+DEFAULT_MODEL=claude-sonnet-5-5
 # or
-DEFAULT_MODEL=gemini-3-pro-preview
+DEFAULT_MODEL=gemini-3.6-flash
 ```
 
 **Mistral AI (chat models with function calling):**
@@ -227,7 +210,7 @@ ANTHROPIC_MODELS=claude-sonnet-4-6,claude-opus-4-6
 GEMINI_DEFAULT_TEMPERATURE=0.7
 GEMINI_DEFAULT_TIMEOUT=120
 GEMINI_DEFAULT_MAX_RETRIES=3
-GEMINI_MODELS=gemini-3-pro-preview,gemini-2.5-flash
+GEMINI_MODELS=gemini-3.1-pro-preview,gemini-3.6-flash
 ```
 
 #### Mistral Settings
@@ -308,9 +291,9 @@ AUTO_CLASSIFIER_MODEL=claude-haiku-4-5-20251001
 
 **Example:**
 ```bash
-PREFERRED_LIGHT_MODELS=gpt-5.4-mini,gemini-2.5-flash-lite
-PREFERRED_STANDARD_MODELS=claude-sonnet-4-20250522,gemini-2.5-flash
-PREFERRED_ADVANCED_MODELS=claude-opus-4-6,gemini-3-pro-preview,gpt-5
+PREFERRED_LIGHT_MODELS=gpt-5.4-nano,gemini-3.5-flash-lite
+PREFERRED_STANDARD_MODELS=claude-sonnet-5-5,gemini-3.6-flash
+PREFERRED_ADVANCED_MODELS=claude-opus-5-5,gemini-3.1-pro-preview,gpt-5.5
 ```
 
 ## Session Storage

{agent_framework_lib-0.8.15 → agent_framework_lib-0.8.17/agent_framework_lib.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agent-framework-lib
-Version: 0.8.15
+Version: 0.8.17
 Summary: A comprehensive Python framework for building and serving conversational AI agents with FastAPI
 Author-email: Sebastian Pavel <sebastian@cinco.ai>, Elliott Girard <elliott.girard@icloud.com>
 Maintainer-email: Sebastian Pavel <sebastian@cinco.ai>
@@ -333,7 +333,7 @@ Pour **déployer** un agent, partez de l'image officielle — Chromium Playwrigh
 sandbox locale déjà en place (sinon l'agent ne peut ni rendre un PNG ni installer quoi que ce soit) :
 
 ```dockerfile
-FROM ghcr.io/komus-ai/agent-framework-runtime:0.8.15
+FROM ghcr.io/komus-ai/agent-framework-runtime:0.8.17
 ```
 
 Détails, publication et visibilité GHCR : [docs/AGENT_RUNTIME_IMAGE.md](docs/AGENT_RUNTIME_IMAGE.md).
@@ -542,9 +542,9 @@ DEFAULT_MODEL_MODE=auto
 AUTO_CLASSIFIER_MODEL=gpt-5.4-nano
 
 # Preferred models per tier (comma-separated, in order of preference)
-PREFERRED_LIGHT_MODELS=gpt-5.4-nano,claude-haiku-4-5-20251001,gemini-2.5-flash-lite
-PREFERRED_STANDARD_MODELS=gpt-5.4-mini,gemini-2.5-flash
-PREFERRED_ADVANCED_MODELS=gpt-5.2,claude-opus-4-6,gemini-2.5-pro
+PREFERRED_LIGHT_MODELS=gpt-5.4-nano,claude-haiku-4-5-20251001,gemini-3.5-flash-lite
+PREFERRED_STANDARD_MODELS=gpt-5.4-mini,gemini-3.6-flash
+PREFERRED_ADVANCED_MODELS=gpt-5.2,claude-opus-4-6,gemini-3.1-pro-preview
 ```
 
 ### API Endpoint
@@ -596,7 +596,7 @@ class MyAgent(LlamaIndexAgent):
 **Runtime Configuration:**
 
 Users can override settings per session via the API or web UI:
-- Model selection (gpt-5.2, claude-sonnet-4-6, gemini-2.5-pro)
+- Model selection (gpt-5.2, claude-sonnet-4-6, gemini-3.1-pro-preview)
 - Temperature (0.0 - 1.0)
 - Max tokens
 - System prompt override

{agent_framework_lib-0.8.15 → agent_framework_lib-0.8.17}/docs/AGENT_RUNTIME_IMAGE.md

@@ -33,7 +33,7 @@ Architectures : `linux/amd64`, `linux/arm64` (VM Mozart Scaleway = arm64).
 ## Utilisation dans un agent
 
 ```dockerfile
-FROM ghcr.io/komus-ai/agent-framework-runtime:0.8.15
+FROM ghcr.io/komus-ai/agent-framework-runtime:0.8.17
 
 WORKDIR /app
 # Deps figées par uv.lock (jamais de résolution à neuf au build : cf. incident anthropic 1.5.0).
@@ -53,7 +53,7 @@ l'agent réinstalle ses deps depuis son lock, l'image fournit le système (Chrom
 
 - Automatique : workflow `.github/workflows/runtime-image.yml` sur tag `v*.*.*`, après attente de la
   version sur PyPI.
-- Manuelle : `gh workflow run runtime-image.yml -f version=0.8.15` (version déjà sur PyPI).
+- Manuelle : `gh workflow run runtime-image.yml -f version=0.8.17` (version déjà sur PyPI).
 - **Visibilité** : un paquet GHCR est privé par défaut. Mozart construit les images d'agents avec le
   CLI Docker du conteneur `mozart-server`, sans identifiants GHCR → le paquet doit être **public**
   (GitHub → Packages → `agent-framework-runtime` → Package settings → Change visibility). À faire une
@@ -62,7 +62,7 @@ l'agent réinstalle ses deps depuis son lock, l'image fournit le système (Chrom
 ## Vérifier une image
 
 ```bash
-docker run --rm ghcr.io/komus-ai/agent-framework-runtime:0.8.15 sh -c '
+docker run --rm ghcr.io/komus-ai/agent-framework-runtime:0.8.17 sh -c '
   whoami; deno --version | head -1; python -c "
 from playwright.sync_api import sync_playwright
 with sync_playwright() as p:

{agent_framework_lib-0.8.15/agent_framework/web → agent_framework_lib-0.8.17}/docs/configuration.md

@@ -108,47 +108,30 @@ MISTRAL_API_KEY=your-mistral-key
 - **Default**: `claude-haiku-4-5-20251001`
 - **Description**: Default model to use when no model is specified in agent configuration. The framework automatically detects the provider from the model name.
 
-**Supported Models:**
+**Supported Models** (checked 29/09/2026 through the lib: chat, streaming and tool call):
+
+**Anthropic:**
+- `claude-opus-5-5`, `claude-opus-5`, `claude-opus-4-8`, `claude-opus-4-7`, `claude-opus-4-6`, `claude-opus-4-5-20251101`
+- `claude-sonnet-5-5`, `claude-sonnet-5`, `claude-sonnet-4-6`
+- `claude-haiku-4-5-20251001`
+- `claude-fable-5-1`, `claude-fable-5`
 
 **OpenAI:**
-- `gpt-5.2` - GPT-5.2 (latest flagship)
-- `gpt-5` - GPT-5
-- `gpt-5.4-mini` - Smaller, fast
-- `gpt-5.4-nano` - Ultra-lightweight, cost-efficient
-- `o1` - O1
-
-**Anthropic (Claude 4 Family - Latest):**
-- `claude-opus-4-6` - Claude Opus 4.6 (most powerful, adaptive thinking, 1M context beta, released February 2026)
-- `claude-opus-4-5-20251101` - Claude Opus 4.5
-- `claude-opus-4-1-20250514` - Claude Opus 4.1 (incremental update to Opus 4)
-- `claude-sonnet-4-6` - Claude Sonnet 4.6 (latest Sonnet, balanced performance)
-- `claude-haiku-4-5-20251001` - Claude Haiku 4.5 (fastest, most cost-efficient)
-
-**Anthropic (Claude 3 Family - Previous Generation):**
-- `claude-3-7-sonnet-20250219` - Claude 3.7 Sonnet
-- `claude-3-5-sonnet-20241022` - Claude 3.5 Sonnet
-- `claude-3-5-haiku-20241022` - Claude 3.5 Haiku
-
-**Google Gemini (Gemini 3 Family - Latest):**
-- `gemini-3-pro-preview` - Gemini 3 Pro (most intelligent, state-of-the-art reasoning)
-- `gemini-3-deep-think` - Gemini 3 Deep Think (advanced reasoning mode)
-
-**Google Gemini (Gemini 2.5 Family):**
-- `gemini-2.5-pro` - Gemini 2.5 Pro (powerful with adaptive thinking)
-- `gemini-2.5-flash` - Gemini 2.5 Flash (fast, well-rounded capabilities)
-- `gemini-2.5-flash-lite` - Gemini 2.5 Flash-Lite (optimized for cost-efficiency)
-
-**Google Gemini (Gemini 2.0 Family):**
-- `gemini-2.0-flash` - Gemini 2.0 Flash (1M token context window)
-- `gemini-2.0-flash-thinking` - Gemini 2.0 Flash Thinking (shows reasoning process)
+- `gpt-6-sol`, `gpt-6-luna`, `gpt-6-astra`, `gpt-5.6-sol`, `gpt-5.6-terra`, `gpt-5.6-luna`, `gpt-5.5-pro`, `gpt-5.4-pro` (served through the OpenAI Responses API)
+- `gpt-5.5`, `gpt-5.4`, `gpt-5.4-mini`, `gpt-5.4-nano`
+
+**Google Gemini:**
+- `gemini-3.8-flash`, `gemini-3.7-flash`, `gemini-3.6-flash`, `gemini-3.5-flash`, `gemini-3-flash-preview`
+- `gemini-3.5-flash-lite`, `gemini-3.1-flash-lite`
+- `gemini-3.1-pro-preview`
 
 **Example:**
 ```bash
 DEFAULT_MODEL=claude-haiku-4-5-20251001
 # or
-DEFAULT_MODEL=claude-sonnet-4-6
+DEFAULT_MODEL=claude-sonnet-5-5
 # or
-DEFAULT_MODEL=gemini-3-pro-preview
+DEFAULT_MODEL=gemini-3.6-flash
 ```
 
 **Mistral AI (chat models with function calling):**
@@ -227,7 +210,7 @@ ANTHROPIC_MODELS=claude-sonnet-4-6,claude-opus-4-6
 GEMINI_DEFAULT_TEMPERATURE=0.7
 GEMINI_DEFAULT_TIMEOUT=120
 GEMINI_DEFAULT_MAX_RETRIES=3
-GEMINI_MODELS=gemini-3-pro-preview,gemini-2.5-flash
+GEMINI_MODELS=gemini-3.1-pro-preview,gemini-3.6-flash
 ```
 
 #### Mistral Settings
@@ -308,9 +291,9 @@ AUTO_CLASSIFIER_MODEL=claude-haiku-4-5-20251001
 
 **Example:**
 ```bash
-PREFERRED_LIGHT_MODELS=gpt-5.4-mini,gemini-2.5-flash-lite
-PREFERRED_STANDARD_MODELS=claude-sonnet-4-20250522,gemini-2.5-flash
-PREFERRED_ADVANCED_MODELS=claude-opus-4-6,gemini-3-pro-preview,gpt-5
+PREFERRED_LIGHT_MODELS=gpt-5.4-nano,gemini-3.5-flash-lite
+PREFERRED_STANDARD_MODELS=claude-sonnet-5-5,gemini-3.6-flash
+PREFERRED_ADVANCED_MODELS=claude-opus-5-5,gemini-3.1-pro-preview,gpt-5.5
 ```
 
 ## Session Storage

{agent_framework_lib-0.8.15 → agent_framework_lib-0.8.17}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "agent-framework-lib"
-version = "0.8.15"
+version = "0.8.17"
 description = "A comprehensive Python framework for building and serving conversational AI agents with FastAPI"
 readme = {file = "README.md", content-type = "text/markdown"}
 license = {text = "MIT"}