pico-cli-ai 0.1.1__tar.gz

pico_cli_ai-0.1.1/.gitignore

@@ -0,0 +1,25 @@
+# Python
+__pycache__/
+*.py[cod]
+*$py.class
+*.egg-info/
+*.egg
+dist/
+build/
+.eggs/
+
+# Virtual environments
+.venv/
+venv/
+env/
+
+# Test / coverage
+.pytest_cache/
+.coverage
+htmlcov/
+
+# IDE / OS
+.idea/
+.vscode/
+.DS_Store
+Thumbs.db

pico_cli_ai-0.1.1/PKG-INFO

@@ -0,0 +1,26 @@
+Metadata-Version: 2.5
+Name: pico-cli-ai
+Version: 0.1.1
+Summary: LLM abstraction and protocol normalisation.
+Project-URL: Homepage, https://github.com/Arya-Ojha/PicoCLI_Learn
+Project-URL: Repository, https://github.com/Arya-Ojha/PicoCLI_Learn
+Project-URL: Issues, https://github.com/Arya-Ojha/PicoCLI_Learn/issues
+Author-email: Arya-Ojha <aryaojha195@gmail.com>
+License-Expression: MIT
+Keywords: ai-agent,anthropic,llm,openai,openrouter
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.12
+Requires-Python: >=3.12
+Requires-Dist: httpx>=0.27
+Requires-Dist: pydantic>=2
+Description-Content-Type: text/markdown
+
+# pico-cli-ai
+
+LLM abstraction and protocol normalisation for `pico-cli`: one unified streaming "AI call" shape plus per-provider adapters (OpenRouter, OpenAI, Anthropic, Gemini, DeepSeek, local Ollama).
+
+See the [pico-cli README](https://github.com/Arya-Ojha/PicoCLI_Learn#readme) for the full picture.

pico_cli_ai-0.1.1/README.md

@@ -0,0 +1,5 @@
+# pico-cli-ai
+
+LLM abstraction and protocol normalisation for `pico-cli`: one unified streaming "AI call" shape plus per-provider adapters (OpenRouter, OpenAI, Anthropic, Gemini, DeepSeek, local Ollama).
+
+See the [pico-cli README](https://github.com/Arya-Ojha/PicoCLI_Learn#readme) for the full picture.

pico_cli_ai-0.1.1/pyproject.toml

@@ -0,0 +1,30 @@
+[project]
+name = "pico-cli-ai"
+version = "0.1.1"
+description = "LLM abstraction and protocol normalisation."
+readme = "README.md"
+license = "MIT"
+authors = [{ name = "Arya-Ojha", email = "aryaojha195@gmail.com" }]
+keywords = ["llm", "ai-agent", "openrouter", "openai", "anthropic"]
+classifiers = [
+    "Development Status :: 4 - Beta",
+    "Intended Audience :: Developers",
+    "License :: OSI Approved :: MIT License",
+    "Operating System :: OS Independent",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.12",
+]
+requires-python = ">=3.12"
+dependencies = ["pydantic>=2", "httpx>=0.27"]
+
+[project.urls]
+Homepage = "https://github.com/Arya-Ojha/PicoCLI_Learn"
+Repository = "https://github.com/Arya-Ojha/PicoCLI_Learn"
+Issues = "https://github.com/Arya-Ojha/PicoCLI_Learn/issues"
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[tool.hatch.build.targets.wheel]
+packages = ["src/pico_ai"]

pico_cli_ai-0.1.1/src/pico_ai/__init__.py

@@ -0,0 +1 @@
+"""pico_ai: LLM abstraction and protocol normalisation."""

pico_cli_ai-0.1.1/src/pico_ai/openrouter.py

@@ -0,0 +1,262 @@
+"""OpenRouter provider: streams chat completions and normalizes to StreamEvents."""
+
+from __future__ import annotations
+
+import asyncio
+import json
+from collections.abc import AsyncIterator
+
+import httpx
+
+from .types import AICallRequest, StreamEvent, ToolCall, Usage
+
+
+class OpenRouterProvider:
+    """Streams chat completions from OpenRouter and normalizes them."""
+
+    provider_id = "openrouter"
+    display_name = "OpenRouter"
+
+    def __init__(
+        self,
+        api_key: str,
+        base_url: str = "https://openrouter.ai/api/v1",
+        client: httpx.AsyncClient | None = None,
+        timeout: httpx.Timeout | None = None,
+        first_token_timeout: float = 60.0,
+    ) -> None:
+        self._api_key = api_key
+        self._base_url = base_url
+        self._client = client
+        # Streaming LLM responses can take a while before the first token; the
+        # httpx default (5s) is far too short, so default to a generous read.
+        self._timeout = timeout or httpx.Timeout(300.0, connect=10.0)
+        # Abort if the model produces no first token within this many seconds;
+        # without it a stalled request looks like an infinite hang.
+        self._first_token_timeout = first_token_timeout
+        # model id -> whether the model supports tool calling (populated by
+        # list_models); unknown models are assumed to support tools.
+        self._tool_support: dict[str, bool] = {}
+
+    async def stream(self, request: AICallRequest) -> AsyncIterator[StreamEvent]:
+        payload = self._build_payload(request)
+        headers = {
+            "Authorization": f"Bearer {self._api_key}",
+            "Content-Type": "application/json",
+        }
+        client = self._client or httpx.AsyncClient(timeout=self._timeout)
+        pending: dict[int, dict] = {}
+        response_cm = client.stream(
+            "POST",
+            f"{self._base_url}/chat/completions",
+            json=payload,
+            headers=headers,
+        )
+        response: httpx.Response | None = None
+        try:
+            # Bound the wait for the response headers plus the first SSE line:
+            # a stalled upstream (e.g. a broken/free model) would otherwise
+            # hang for the full 300s read timeout looking like an infinite
+            # spin. Once tokens are flowing, the httpx read timeout governs
+            # inter-chunk gaps.
+            try:
+                async with asyncio.timeout(self._first_token_timeout):
+                    response = await response_cm.__aenter__()
+                    if response.status_code >= 400:
+                        # Surface the API's error body: raise_for_status()
+                        # alone hides the actual reason (e.g. unsupported
+                        # tool calling for the chosen model).
+                        body = (await response.aread()).decode(
+                            errors="replace"
+                        )
+                        raise RuntimeError(
+                            f"OpenRouter error {response.status_code} "
+                            f"for model '{request.model}': "
+                            f"{body[:500]}"
+                        )
+                    lines = response.aiter_lines()
+                    first_line = await anext(lines, None)
+            except TimeoutError as exc:
+                raise RuntimeError(
+                    f"no response from model within "
+                    f"{self._first_token_timeout:g}s "
+                    f"(first-token timeout); try another model"
+                ) from exc
+            if first_line is not None:
+                async for event in self._emit_lines(
+                    self._prepend(first_line, lines), pending
+                ):
+                    yield event
+        finally:
+            if response is not None:
+                await response_cm.__aexit__(None, None, None)
+            if self._client is None:
+                await client.aclose()
+
+    async def _emit_lines(
+        self, lines: AsyncIterator[str], pending: dict[int, dict]
+    ) -> AsyncIterator[StreamEvent]:
+        """Parse SSE lines and yield normalized stream events."""
+        async for line in lines:
+            if not line.startswith("data:"):
+                continue
+            data = line[len("data:") :].strip()
+            if data == "[DONE]":
+                return
+            for event in self._parse_chunk(json.loads(data), pending):
+                yield event
+
+    @staticmethod
+    async def _prepend(
+        first: str, rest: AsyncIterator[str]
+    ) -> AsyncIterator[str]:
+        yield first
+        async for item in rest:
+            yield item
+
+    async def list_models(self) -> list[dict]:
+        """Return the available models from OpenRouter.
+
+        Each entry is a dict with keys ``id``, ``name`` and ``is_free``.
+        """
+        headers = {}
+        if self._api_key:
+            headers["Authorization"] = f"Bearer {self._api_key}"
+        client = self._client or httpx.AsyncClient(timeout=self._timeout)
+        try:
+            response = await client.get(
+                f"{self._base_url}/models",
+                headers=headers,
+            )
+            response.raise_for_status()
+            data = response.json().get("data", [])
+        finally:
+            if self._client is None:
+                await client.aclose()
+        models: list[dict] = []
+        for entry in data:
+            pricing = entry.get("pricing") or {}
+            is_free = (
+                str(pricing.get("prompt", "1")).strip() in ("0", "0.0", "-1")
+                and str(pricing.get("completion", "1")).strip() in ("0", "0.0", "-1")
+            )
+            supported = entry.get("supported_parameters") or []
+            supports_tools = any(
+                p in supported for p in ("tools", "tool_choice")
+            )
+            model_id = entry.get("id", "")
+            # Remember tool support so stream() can omit tools for models
+            # that reject them (OpenRouter answers 400 otherwise).
+            self._tool_support[model_id] = supports_tools
+            models.append(
+                {
+                    "id": model_id,
+                    "name": entry.get("name") or model_id,
+                    "is_free": is_free,
+                    "supports_tools": supports_tools,
+                }
+            )
+        return models
+
+    def _build_payload(self, request: AICallRequest) -> dict:
+        messages: list[dict] = []
+        if request.system:
+            messages.append({"role": "system", "content": request.system})
+        for m in request.messages:
+            msg: dict = {"role": m.role, "content": m.content}
+            if m.tool_calls:
+                msg["tool_calls"] = [
+                    {
+                        "id": tc.id,
+                        "type": "function",
+                        "function": {
+                            "name": tc.name,
+                            "arguments": json.dumps(tc.arguments),
+                        },
+                    }
+                    for tc in m.tool_calls
+                ]
+                # An assistant turn that only made tool calls has no text;
+                # send null content rather than an empty string.
+                if not m.content:
+                    msg["content"] = None
+            if m.tool_call_id is not None:
+                msg["tool_call_id"] = m.tool_call_id
+            if m.name is not None:
+                msg["name"] = m.name
+            messages.append(msg)
+        payload: dict = {
+            "model": request.model,
+            "messages": messages,
+            "stream": True,
+            "stream_options": {"include_usage": True},
+        }
+        if request.tools and self._tool_support.get(request.model, True):
+            payload["tools"] = [
+                {
+                    "type": "function",
+                    "function": {
+                        "name": t.name,
+                        "description": t.description,
+                        "parameters": t.input_schema,
+                    },
+                }
+                for t in request.tools
+            ]
+        return payload
+
+    def _parse_chunk(self, chunk: dict, pending: dict[int, dict]) -> list[StreamEvent]:
+        events: list[StreamEvent] = []
+        for choice in chunk.get("choices") or []:
+            delta = choice.get("delta") or {}
+            content = delta.get("content")
+            if content:
+                events.append(StreamEvent(kind="text", text=content))
+            reasoning = delta.get("reasoning") or delta.get("reasoning_content")
+            if reasoning:
+                events.append(StreamEvent(kind="thinking", thinking=reasoning))
+            for tc in delta.get("tool_calls") or []:
+                idx = tc.get("index", 0)
+                entry = pending.setdefault(idx, {"id": "", "name": "", "args": ""})
+                if tc.get("id"):
+                    entry["id"] = tc["id"]
+                fn = tc.get("function") or {}
+                if fn.get("name"):
+                    entry["name"] = fn["name"]
+                if fn.get("arguments"):
+                    entry["args"] += fn["arguments"]
+            if choice.get("finish_reason") == "tool_calls":
+                for idx in sorted(pending):
+                    entry = pending[idx]
+                    events.append(
+                        StreamEvent(
+                            kind="tool_call",
+                            tool_call=ToolCall(
+                                id=entry["id"],
+                                name=entry["name"],
+                                arguments=self._parse_args(entry["args"]),
+                            ),
+                        )
+                    )
+                pending.clear()
+        usage = chunk.get("usage")
+        if usage:
+            events.append(
+                StreamEvent(
+                    kind="usage",
+                    usage=Usage(
+                        input_tokens=usage.get("prompt_tokens", 0),
+                        output_tokens=usage.get("completion_tokens", 0),
+                        total_tokens=usage.get("total_tokens", 0),
+                    ),
+                )
+            )
+        return events
+
+    @staticmethod
+    def _parse_args(raw: str) -> dict:
+        try:
+            parsed = json.loads(raw)
+            return parsed if isinstance(parsed, dict) else {}
+        except json.JSONDecodeError:
+            return {}

pico_cli_ai-0.1.1/src/pico_ai/provider.py

@@ -0,0 +1,16 @@
+"""The provider seam: a single async streaming interface every backend implements."""
+
+from __future__ import annotations
+
+from collections.abc import AsyncIterator
+from typing import Protocol
+
+from .types import AICallRequest, StreamEvent
+
+
+class Provider(Protocol):
+    """A backend that turns a request into a stream of normalized events."""
+
+    def stream(self, request: AICallRequest) -> AsyncIterator[StreamEvent]:
+        """Yield normalized stream events for a request."""
+        ...

pico_cli_ai-0.1.1/src/pico_ai/providers/__init__.py

@@ -0,0 +1,86 @@
+"""Provider registry: the host-owned list of available backends (ADR-0004).
+
+Adding a provider means adding one module next to this one (converting that
+provider's wire format to the app's default ``StreamEvent`` shape) plus one
+line in ``PROVIDERS`` below. This is a hardcoded list, not a plugin API —
+the host keeps sovereignty over which backends exist.
+"""
+
+from __future__ import annotations
+
+from typing import Any
+
+from pico_ai.openrouter import OpenRouterProvider
+
+from . import anthropic, deepseek, gemini, ollama
+from . import openai as openai_provider
+from .spec import FieldSpec, ProviderSpec
+
+OPENROUTER_DEFAULT_BASE_URL = "https://openrouter.ai/api/v1"
+
+
+def _create_openrouter(config: dict[str, str]) -> OpenRouterProvider:
+    return OpenRouterProvider(
+        api_key=config.get("api_key", ""),
+        base_url=config.get("base_url", "") or OPENROUTER_DEFAULT_BASE_URL,
+    )
+
+
+OPENROUTER_SPEC = ProviderSpec(
+    id="openrouter",
+    display_name="OpenRouter",
+    description="OpenRouter gateway (many models, one key).",
+    fields=(
+        FieldSpec(
+            key="api_key",
+            label="API key",
+            secret=True,
+            required=True,
+            env_var="OPENROUTER_API_KEY",
+            placeholder="sk-or-v1-...",
+        ),
+        FieldSpec(
+            key="base_url",
+            label="Base URL",
+            default=OPENROUTER_DEFAULT_BASE_URL,
+            placeholder=OPENROUTER_DEFAULT_BASE_URL,
+        ),
+        FieldSpec(key="model", label="Default model", default="openrouter/free"),
+    ),
+    default_model="openrouter/free",
+    create=_create_openrouter,
+)
+
+PROVIDERS: dict[str, ProviderSpec] = {
+    OPENROUTER_SPEC.id: OPENROUTER_SPEC,
+    openai_provider.SPEC.id: openai_provider.SPEC,
+    anthropic.SPEC.id: anthropic.SPEC,
+    gemini.SPEC.id: gemini.SPEC,
+    deepseek.SPEC.id: deepseek.SPEC,
+    ollama.SPEC.id: ollama.SPEC,
+}
+
+
+def provider_ids() -> list[str]:
+    """Return registered provider ids in picker order."""
+    return list(PROVIDERS)
+
+
+def get_spec(provider_id: str) -> ProviderSpec:
+    """Return the spec for ``provider_id`` (``KeyError`` when unknown)."""
+    return PROVIDERS[provider_id]
+
+
+def create_provider(provider_id: str, config: dict[str, str]) -> Any:
+    """Build the adapter for ``provider_id`` from a config dict."""
+    return get_spec(provider_id).create(config)
+
+
+__all__ = [
+    "PROVIDERS",
+    "ProviderSpec",
+    "FieldSpec",
+    "provider_ids",
+    "get_spec",
+    "create_provider",
+]