feat: add ollama provider and new openai models (gpt-oss:20b and gpt-oss:120b) (#30)

runemalm · web-flow · commit 6382c7794ee5 · 2025-08-10T15:04:53.000+02:00
diff --git a/README.md b/README.md
@@ -27,14 +27,6 @@
 
 ---
 
-> ⚠️ **Alpha Release**
->
-> Codius is currently in alpha.  
-> Features may be incomplete, unstable, or subject to change.  
-> Use with care in production environments — and help shape it by providing feedback or contributing!
-
----
-
 ## Why Codius?
 
 > Free domain-driven designers from boilerplate and let them focus on modeling.
diff --git a/poetry.lock b/poetry.lock
diff --git a/pyproject.toml b/pyproject.toml
@@ -26,15 +26,15 @@ dependencies = [
     "jinja2",
     "langgraph",
     "langchain-openai",
-    "py-dependency-injection >=1.0.0b3,<2.0.0",
     "python-dateutil",
     "python-dotenv",
     "pyyaml",
     "prompt-toolkit",
     "rich",
     "tree-sitter==0.23.2",
     "tree-sitter-c-sharp",
-    "typing-extensions"
+    "typing-extensions",
+    "py-dependency-injection (>=1.0.0rc2,<2.0.0)",
 ]
 
 [tool.poetry.group.dev]
diff --git a/src/codius/di.py b/src/codius/di.py
@@ -3,9 +3,11 @@
 from dependency_injection.container import DependencyContainer
 
 from codius.domain.model.config.config import Config
+from codius.domain.model.config.llm_provider import LlmProvider
 from codius.domain.model.port.llm_port import LlmPort
 from codius.domain.services.config_service import ConfigService
 from codius.domain.services.session_service import SessionService
+from codius.infrastructure.adapter.llm.ollama.ollama_llm_adapter import OllamaLlmAdapter
 
 from codius.infrastructure.adapter.llm.openai.openai_llm_adapter import OpenAiLlmAdapter
 from codius.infrastructure.repository.session_repository import SessionRepository
@@ -52,4 +54,16 @@ def register_services(config: Config, args: argparse.Namespace):
     container.register_scoped(OpenDddConventionService)
     container.register_scoped(TreeSitterService)
     container.register_scoped(LlmService)
-    container.register_scoped(LlmPort, OpenAiLlmAdapter)
+
+    adapter_by_provider = {
+        LlmProvider.OPENAI: OpenAiLlmAdapter,
+        LlmProvider.OLLAMA: OllamaLlmAdapter,
+    }
+
+    provider = config.llm.provider
+    adapter_cls = adapter_by_provider.get(provider)
+
+    if adapter_cls is None:
+        raise RuntimeError(f"Unsupported LLM provider in config: {provider}")
+
+    container.register_scoped(LlmPort, adapter_cls)
diff --git a/src/codius/domain/model/config/config.py b/src/codius/domain/model/config/config.py
@@ -9,6 +9,14 @@
         "openai": {
             "model": "gpt-4o",
             "api_key": "sk-... # Replace with your OpenAI API key"
+        },
+        "anthropic": {
+            "model": "claude-3-opus",
+            "api_key": "xyz-... # Replace with your Anthropic API key"
+        },
+        "ollama": {
+            "base_url": "http://localhost:11434",
+            "model": "gpt-oss:20b"
         }
     },
     "approval_mode": ApprovalMode.SUGGEST.value,
diff --git a/src/codius/domain/model/config/llm_provider.py b/src/codius/domain/model/config/llm_provider.py
@@ -7,3 +7,4 @@ class LlmProvider(str, Enum):
     GOOGLE = "google"
     MISTRAL = "mistral"
     GROQ = "groq"
+    OLLAMA = "ollama"
diff --git a/src/codius/domain/model/config/ollama/__init__.py b/src/codius/domain/model/config/ollama/__init__.py
diff --git a/src/codius/domain/model/config/ollama/ollama_llm_model.py b/src/codius/domain/model/config/ollama/ollama_llm_model.py
@@ -0,0 +1,8 @@
+from enum import Enum
+
+
+class OllamaModel(str, Enum):
+    GPT_OSS_20B = "gpt-oss:20b"
+    GPT_OSS_120B = "gpt-oss:120b"
+    LLAMA_31_8B = "llama3.1:8b-instruct-q4_K_M"
+    MISTRAL_7B = "mistral:7b"
diff --git a/src/codius/domain/model/intents/intent_type.py b/src/codius/domain/model/intents/intent_type.py
@@ -1,5 +1,8 @@
 from enum import Enum
 
+from codius.infrastructure.services.code_scanner.model.building_block_type import \
+    BuildingBlockType
+
 
 class IntentType(str, Enum):
     ADD_AGGREGATE = "add_aggregate"
@@ -22,3 +25,17 @@ class IntentType(str, Enum):
     REMOVE_REPOSITORY_METHOD = "remove_repository_method"
 
     UNSURE = "unsure"
+
+    @property
+    def building_block(self) -> BuildingBlockType:
+        if self.name.startswith("ADD_AGGREGATE") or self.name.startswith(
+            "REMOVE_AGGREGATE"):
+            return BuildingBlockType.AGGREGATE_ROOT
+        elif self.name.startswith("ADD_VALUE_OBJECT") or self.name.startswith(
+            "REMOVE_VALUE_OBJECT"):
+            return BuildingBlockType.VALUE_OBJECT
+        elif self.name.startswith("ADD_REPOSITORY") or self.name.startswith(
+            "REMOVE_REPOSITORY"):
+            return BuildingBlockType.REPOSITORY
+        else:
+            raise NotImplementedError(f"Building block not mapped for intent {self}")
diff --git a/src/codius/domain/model/prompts/distill_intent_prompt.py b/src/codius/domain/model/prompts/distill_intent_prompt.py
@@ -22,6 +22,8 @@
 from codius.domain.model.intents.value_object.add_value_object_property_intent import AddValueObjectPropertyIntent
 from codius.domain.model.intents.value_object.remove_value_object_property_intent import RemoveValueObjectPropertyIntent
 from codius.domain.model.intents.repository.add_repository_intent import AddRepositoryIntent
+from codius.infrastructure.services.code_scanner.model.building_block_type import \
+    BuildingBlockType
 
 
 @dataclass(frozen=True)
@@ -49,6 +51,14 @@ def as_prompt(self) -> str:
             RemoveRepositoryMethodIntent,
         ]
 
+        all_blocks = {bb.value for bb in BuildingBlockType}
+        supported_blocks = {intent.building_block.value for intent in IntentType if
+                            intent != IntentType.UNSURE}
+        unsupported_blocks = all_blocks - supported_blocks
+
+        supported_blocks_text = "\n".join(f"- {b}" for b in sorted(supported_blocks))
+        unsupported_blocks_text = "\n".join(f"- {b}" for b in sorted(unsupported_blocks))
+
         example_blocks = "\n".join(
             f"### {cls.intent.value}\n```json\n{cls.to_example_json()}\n```"
             for cls in example_intents
@@ -88,6 +98,29 @@ def as_prompt(self) -> str:
 **Database Providers**:
 {database_text}
 
+---
+
+### Known DDD Building Blocks
+
+These are the known building blocks in Domain-Driven Design:
+
+**Supported:**
+{supported_blocks_text}
+
+**Not yet supported:**
+{unsupported_blocks_text}
+
+If the user refers to a known DDD building block that is **not yet supported**, return:
+
+```json
+{{ "intent": "unsupported", "building_block": "<block_name>" }}
+```
+
+If the user's intent is unclear or unsupported, respond only with:
+
+```json
+{{ "intent": "unsure" }}
+
 ### Instructions
 
 - Break complex modeling instructions into small, **granular intents**.
@@ -109,9 +142,4 @@ def as_prompt(self) -> str:
 ### Examples:
 
 {example_blocks}
-
-If the user's intent is unclear or unsupported, respond only with:
-
-```json
-{{ "intent": "unsure" }}
 """
diff --git a/src/codius/domain/services/config_service.py b/src/codius/domain/services/config_service.py
@@ -7,10 +7,12 @@
 from codius.domain.model.config.approval_mode import ApprovalMode
 from codius.domain.model.config.config import Config
 from codius.domain.model.config.llm_provider import LlmProvider
+from codius.domain.model.config.ollama.ollama_llm_model import OllamaModel
 from codius.domain.model.config.openai.openai_llm_model import OpenAiModel
 
 from codius.infrastructure.adapter.llm.anthropic.anthropic_config import AnthropicConfig
 from codius.infrastructure.adapter.llm.llm_config import LlmConfig
+from codius.infrastructure.adapter.llm.ollama.ollama_config import OllamaConfig
 from codius.infrastructure.adapter.llm.openai.openai_config import OpenAiConfig
 from codius.infrastructure.services.project_metadata_service import ProjectMetadataService
 
@@ -55,20 +57,27 @@ def parse_structured(cls, raw: dict) -> Config:
 
         llm_config = LlmConfig(provider=provider)
 
-        openai = llm_section.get("openai", {})
-        raw_model = openai.get("model") or "gpt-4o"
-        llm_config.openai = OpenAiConfig(
-            model=OpenAiModel(raw_model),
-            api_key=openai.get("api_key", "")
-        )
-
         anthropic = llm_section.get("anthropic", {})
         raw_model = anthropic.get("model") or "claude-3-opus"
         llm_config.anthropic = AnthropicConfig(
             model=AnthropicModel(raw_model),
             api_key=anthropic.get("api_key", "")
         )
 
+        ollama = llm_section.get("ollama", {})
+        raw_model = ollama.get("model") or OllamaModel.GPT_OSS_20B.value
+        llm_config.ollama = OllamaConfig(
+            model=OllamaModel(raw_model),
+            server_url=ollama.get("server_url", "")
+        )
+
+        openai = llm_section.get("openai", {})
+        raw_model = openai.get("model") or "gpt-4o"
+        llm_config.openai = OpenAiConfig(
+            model=OpenAiModel(raw_model),
+            api_key=openai.get("api_key", "")
+        )
+
         log_level = raw.get("log_level", "warning").lower()
         allowed_levels = {"debug", "info", "warning", "error", "critical"}
         if log_level not in allowed_levels:
diff --git a/src/codius/graph/nodes/handle_unsupported_intent.py b/src/codius/graph/nodes/handle_unsupported_intent.py
@@ -0,0 +1,34 @@
+from typing import cast
+
+
+def handle_unsupported_intent(state: dict) -> dict:
+    from codius.domain.model.intents.intent_type import IntentType
+
+    unsupported_blocks = {
+        intent.get("building_block")
+        for intent in state.get("intent", [])
+        if intent.get("intent") == "unsupported"
+    }
+    block_list = ", ".join(sorted(unsupported_blocks))
+
+    # Dynamically get supported building blocks
+    supported_blocks = {
+        intent.building_block.value
+        for intent in IntentType
+        if intent != IntentType.UNSURE
+    }
+    supported_blocks_list = "\n".join(f"- {b}" for b in sorted(supported_blocks))
+
+    state["final_output"] = (
+        f"⚠️ The assistant understood your request, but the following building block(s) aren't supported yet:\n"
+        f"- {block_list}\n\n"
+        "We're shipping updates frequently, so it might already be available in a newer version.\n\n"
+        "👉 Try updating Codius:\n"
+        "```bash\npip install --upgrade codius\n```\n"
+        "Then rerun your request.\n\n"
+        "In the meantime, you can try working with one of the currently supported building blocks:\n"
+        f"{supported_blocks_list}\n\n"
+        "💡 Need help or want to request this feature? Let us know on GitHub!"
+    )
+
+    return state
diff --git a/src/codius/graph/routers/intent_router.py b/src/codius/graph/routers/intent_router.py
@@ -16,6 +16,8 @@ def route_by_intent(state: dict) -> str:
 
             if intent_type == "error":
                 return "error"
+            elif intent_type == "unsupported":
+                return "unsupported"
             elif intent_type and intent_type not in {"none", "greeting", "unsure"}:
                 return "valid"
 
diff --git a/src/codius/infrastructure/adapter/llm/llm_config.py b/src/codius/infrastructure/adapter/llm/llm_config.py
@@ -6,6 +6,7 @@
 from codius.infrastructure.adapter.llm.google.google_config import GoogleConfig
 from codius.infrastructure.adapter.llm.groq.groq_config import GroqConfig
 from codius.infrastructure.adapter.llm.mistral.mistral_config import MistralConfig
+from codius.infrastructure.adapter.llm.ollama.ollama_config import OllamaConfig
 from codius.infrastructure.adapter.llm.openai.openai_config import OpenAiConfig
 
 
@@ -17,3 +18,4 @@ class LlmConfig:
     google: Optional[GoogleConfig] = None
     mistral: Optional[MistralConfig] = None
     groq: Optional[GroqConfig] = None
+    ollama: Optional[OllamaConfig] = None
diff --git a/src/codius/infrastructure/adapter/llm/ollama/__init__.py b/src/codius/infrastructure/adapter/llm/ollama/__init__.py
diff --git a/src/codius/infrastructure/adapter/llm/ollama/ollama_config.py b/src/codius/infrastructure/adapter/llm/ollama/ollama_config.py
@@ -0,0 +1,7 @@
+from dataclasses import dataclass
+
+
+@dataclass
+class OllamaConfig:
+    server_url: str
+    model: str
diff --git a/src/codius/infrastructure/adapter/llm/ollama/ollama_llm_adapter.py b/src/codius/infrastructure/adapter/llm/ollama/ollama_llm_adapter.py
@@ -0,0 +1,48 @@
+import logging, time, requests
+from codius.domain.model.config.config import Config
+from codius.domain.model.port.llm_port import LlmPort
+
+logger = logging.getLogger(__name__)
+
+class OllamaLlmAdapter(LlmPort):
+    def __init__(self, config: Config):
+        cfg = config.llm.ollama
+        if not cfg:
+            raise RuntimeError("Ollama config not found. Add `llm.ollama` in config.")
+        self.base_url = cfg.server_url.rstrip("/")
+        self.model = cfg.model
+        self.options = {}
+        self.session = requests.Session()
+        self.timeout = 300
+
+    def call_prompt(self, prompt: str) -> str:
+        start = time.time()
+        payload = {
+            "model": self.model,
+            "prompt": prompt,
+            "options": self.options,
+            "stream": False,
+        }
+        r = self.session.post(f"{self.base_url}/api/generate", json=payload, timeout=self.timeout)
+        r.raise_for_status()
+        data = r.json()
+        content = data.get("response", "")
+        logger.debug("Ollama /generate returned %d chars in %d ms", len(content), int((time.time()-start)*1000))
+        return content
+
+    def call_chat(self, messages: list[dict]) -> str:
+        # messages format: [{"role": "system|user|assistant", "content": "..."}, ...]
+        start = time.time()
+        payload = {
+            "model": self.model,
+            "messages": messages,
+            "options": self.options,
+            "stream": False,
+        }
+        r = self.session.post(f"{self.base_url}/api/chat", json=payload, timeout=self.timeout)
+        r.raise_for_status()
+        data = r.json()
+        # final message is in data["message"]["content"] per Ollama docs
+        msg = (data.get("message") or {}).get("content", "")
+        logger.debug("Ollama /chat returned %d chars in %d ms", len(msg), int((time.time()-start)*1000))
+        return msg
diff --git a/src/codius/infrastructure/services/graph_service.py b/src/codius/infrastructure/services/graph_service.py
diff --git a/src/codius/ui/commands/slash_commands.py b/src/codius/ui/commands/slash_commands.py
