synth-ai 0.2.14__py3-none-any.whl → 0.4.4__py3-none-any.whl

synth_ai/cli/agents/codex.py

@@ -0,0 +1,265 @@
+import os
+import subprocess
+from pathlib import Path
+from typing import Any
+
+import click
+
+from synth_ai.cli.lib.agents import write_agents_md
+from synth_ai.cli.lib.bin import install_bin, verify_bin
+from synth_ai.cli.lib.env import resolve_env_var
+from synth_ai.core.env import get_backend_from_env
+from synth_ai.core.paths import get_bin_path
+from synth_ai.core.urls import BACKEND_URL_SYNTH_RESEARCH_OPENAI
+from synth_ai.data.enums import SYNTH_MODEL_NAMES
+
+try:
+    import tomllib  # Python 3.11+
+except ImportError:
+    try:
+        import tomli as tomllib  # type: ignore
+    except ImportError:
+        tomllib = None  # type: ignore
+
+
+def _load_session_config(config_path: Path | None = None) -> dict[str, Any]:
+    """Load session configuration from TOML file.
+    
+    Looks for [session] section with:
+    - limit_cost_usd: float (default: 20.0)
+    - limit_tokens: int | None
+    - limit_gpu_hours: float | None
+    
+    Returns dict with session limits.
+    """
+    if config_path is None:
+        # Look for codex.toml or synth.toml in current directory
+        for name in ["codex.toml", "synth.toml"]:
+            path = Path(name)
+            if path.exists():
+                config_path = path
+                break
+    
+    if config_path is None or not config_path.exists():
+        # Default: $20 cost limit
+        return {"limit_cost_usd": 20.0}
+    
+    if tomllib is None:
+        return {"limit_cost_usd": 20.0}
+    
+    try:
+        with config_path.open("rb") as f:
+            config = tomllib.load(f)
+        
+        session_config = config.get("session", {})
+        result: dict[str, Any] = {}
+        
+        # Default cost limit is $20 if not specified
+        result["limit_cost_usd"] = session_config.get("limit_cost_usd", 20.0)
+        
+        if "limit_tokens" in session_config:
+            result["limit_tokens"] = session_config["limit_tokens"]
+        if "limit_gpu_hours" in session_config:
+            result["limit_gpu_hours"] = session_config["limit_gpu_hours"]
+        
+        return result
+    except Exception as e:
+        click.echo(f"Warning: Failed to load session config from {config_path}: {e}", err=True)
+        return {"limit_cost_usd": 20.0}
+
+
+@click.command("codex")
+@click.option(
+    "--model",
+    "model_name",
+    type=str,
+    default=None
+)
+@click.option(
+    "--force",
+    is_flag=True,
+    help="Prompt for API keys even if cached values exist."
+)
+@click.option(
+    "--url",
+    "override_url",
+    type=str,
+    default=None,
+)
+@click.option(
+    "--wire-api",
+    "wire_api",
+    type=click.Choice(["chat", "responses"], case_sensitive=False),
+    default=None,
+    help="API wire format: 'chat' for Chat Completions, 'responses' for Responses API"
+)
+@click.option(
+    "--config",
+    "config_path",
+    type=click.Path(exists=True, path_type=Path),
+    default=None,
+    help="Path to TOML config file (default: codex.toml or synth.toml in current directory)"
+)
+def codex_cmd(
+    model_name: str | None = None,
+    force: bool = False,
+    override_url: str | None = None,
+    wire_api: str | None = None,
+    config_path: Path | None = None,
+) -> None:
+
+    while True:
+        bin_path = get_bin_path("codex")
+        if bin_path:
+            break
+        if not install_bin(
+            "Codex",
+            [
+                "brew install codex",
+                "npm install -g @openai/codex"
+            ]
+        ):
+            print("Failed to find your installed Codex")
+            print("Please install from: https://developers.openai.com/codex/cli/")
+            return
+    print(f"Using Codex at {bin_path}")
+
+    if not verify_bin(bin_path):
+        print("Failed to verify Codex is runnable")
+        return
+    
+    write_agents_md()
+    env = os.environ.copy()
+    override_args = []
+    
+    # Load session config from TOML
+    session_config = _load_session_config(config_path)
+    session_limit_cost = session_config.get("limit_cost_usd")
+    session_limit_tokens = session_config.get("limit_tokens")
+    session_limit_gpu_hours = session_config.get("limit_gpu_hours")
+    
+    # Get API key and base URL for session creation
+    api_key = resolve_env_var("SYNTH_API_KEY", override_process_env=force)
+    if override_url:
+        base_url = override_url.rstrip("/")
+        if base_url.endswith("/api"):
+            base_url = base_url[:-4]
+    else:
+        base_url, _ = get_backend_from_env()
+    
+    # Create agent session with limits from config (default: $20 cost limit)
+    session_id: str | None = None
+    if session_limit_tokens or session_limit_cost or session_limit_gpu_hours:
+        try:
+            import asyncio
+            
+            # Create session - org_id will be fetched from backend /me endpoint
+            async def create_session():
+                from synth_ai.cli.local.session import AgentSessionClient
+                
+                client = AgentSessionClient(f"{base_url}/api", api_key)
+                
+                limits = []
+                if session_limit_tokens:
+                    limits.append({
+                        "limit_type": "hard",
+                        "metric_type": "tokens",
+                        "limit_value": float(session_limit_tokens),
+                    })
+                if session_limit_cost:
+                    limits.append({
+                        "limit_type": "hard",
+                        "metric_type": "cost_usd",
+                        "limit_value": float(session_limit_cost),
+                    })
+                if session_limit_gpu_hours:
+                    limits.append({
+                        "limit_type": "hard",
+                        "metric_type": "gpu_hours",
+                        "limit_value": float(session_limit_gpu_hours),
+                    })
+                
+                # org_id will be automatically fetched from /api/v1/me endpoint
+                session = await client.create(
+                    org_id=None,  # Will be fetched from backend
+                    limits=limits,
+                    tracing_session_id=None,
+                    session_type="codex_agent",
+                )
+                return session.session_id
+            
+            session_id = asyncio.run(create_session())
+            env["SYNTH_SESSION_ID"] = session_id
+            # Note: Codex CLI doesn't directly support custom headers
+            # Users may need to configure Codex to pass X-Session-ID header
+            # For now, session is created but not automatically passed to API calls
+            click.echo(f"✓ Created agent session: {session_id}")
+            click.echo("  Note: Set X-Session-ID header in Codex config to use this session")
+            if session_limit_tokens:
+                click.echo(f"  Token limit: {session_limit_tokens:,}")
+            if session_limit_cost:
+                click.echo(f"  Cost limit: ${session_limit_cost:.2f}")
+            if session_limit_gpu_hours:
+                click.echo(f"  GPU hours limit: {session_limit_gpu_hours:.2f}")
+        except Exception as e:
+            print(f"Warning: Failed to create agent session: {e}")
+            print("Continuing without session limits...")
+    
+    if model_name is not None:
+        if model_name not in SYNTH_MODEL_NAMES:
+            raise ValueError(f"model_name={model_name} is invalid. Valid values for model_name: {SYNTH_MODEL_NAMES}")
+        if override_url:
+            url = override_url
+            print("Using override URL:", url)
+        else:
+            url = BACKEND_URL_SYNTH_RESEARCH_OPENAI
+        
+        # Determine wire_api based on URL or explicit option
+        # If URL contains "/responses", default to responses API
+        # Otherwise, use explicit wire_api option or default to chat
+        if wire_api is None:
+            wire_api = "responses" if "/responses" in url or url.endswith("/responses") else "chat"
+        
+        # Build provider config with wire_api
+        provider_config_parts = [
+            'name="Synth"',
+            f'base_url="{url}"',
+            'env_key="OPENAI_API_KEY"',
+            f'wire_api="{wire_api}"'
+        ]
+        provider_config = "{" + ",".join(provider_config_parts) + "}"
+        
+        config_overrides = [
+            f"model_providers.synth={provider_config}",
+            'model_provider="synth"',
+            f'default_model="{model_name}"'
+        ]
+        override_args = [arg for override in config_overrides for arg in ("-c", override)]
+        env["OPENAI_API_KEY"] = resolve_env_var("SYNTH_API_KEY", override_process_env=force)
+        env["SYNTH_API_KEY"] = env["OPENAI_API_KEY"]
+        print(f"Configured with wire_api={wire_api}")
+    
+    try:
+        cmd = ["codex"]
+        if model_name is not None:
+            cmd.extend(["-m", model_name])
+        cmd.extend(override_args)
+        print(" ".join(cmd))
+        subprocess.run(cmd, check=True, env=env)
+    except subprocess.CalledProcessError:
+        print("Failed to run Codex")
+    finally:
+        # End session if created
+        if session_id is not None:
+            try:
+                # Assign to local variable to help type checker
+                final_session_id: str = session_id
+                async def end_session():
+                    from synth_ai.cli.local.session import AgentSessionClient
+                    client = AgentSessionClient(f"{base_url}/api", api_key)
+                    await client.end(final_session_id)
+                    click.echo(f"✓ Ended agent session: {final_session_id}")
+                
+                asyncio.run(end_session())
+            except Exception as e:
+                click.echo(f"Warning: Failed to end agent session: {e}", err=True)

synth_ai/cli/agents/opencode.py

@@ -0,0 +1,253 @@
+import subprocess
+from pathlib import Path
+from typing import Any
+
+import click
+
+from synth_ai.cli.lib.agents import write_agents_md
+from synth_ai.cli.lib.bin import install_bin, verify_bin
+from synth_ai.cli.lib.env import resolve_env_var
+from synth_ai.core.env import get_backend_from_env
+from synth_ai.core.json import create_and_write_json, load_json_to_dict
+from synth_ai.core.paths import get_bin_path
+from synth_ai.core.urls import BACKEND_URL_SYNTH_RESEARCH_BASE
+from synth_ai.data.enums import SYNTH_MODEL_NAMES
+
+CONFIG_PATH = Path.home() / ".config" / "opencode" / "opencode.json"
+AUTH_PATH = Path.home() / ".local" / "share" / "opencode" / "auth.json"
+SYNTH_PROVIDER_ID = "synth"
+
+try:
+    import tomllib  # Python 3.11+
+except ImportError:
+    try:
+        import tomli as tomllib  # type: ignore
+    except ImportError:
+        tomllib = None  # type: ignore
+
+
+def _load_session_config(config_path: Path | None = None) -> dict[str, Any]:
+    """Load session configuration from TOML file.
+    
+    Looks for [session] section with:
+    - limit_cost_usd: float (default: 20.0)
+    - limit_tokens: int | None
+    - limit_gpu_hours: float | None
+    
+    Returns dict with session limits.
+    """
+    if config_path is None:
+        # Look for opencode.toml or synth.toml in current directory
+        for name in ["opencode.toml", "synth.toml"]:
+            path = Path(name)
+            if path.exists():
+                config_path = path
+                break
+    
+    if config_path is None or not config_path.exists():
+        # Default: $20 cost limit
+        return {"limit_cost_usd": 20.0}
+    
+    if tomllib is None:
+        return {"limit_cost_usd": 20.0}
+    
+    try:
+        with config_path.open("rb") as f:
+            config = tomllib.load(f)
+        
+        session_config = config.get("session", {})
+        result: dict[str, Any] = {}
+        
+        # Default cost limit is $20 if not specified
+        result["limit_cost_usd"] = session_config.get("limit_cost_usd", 20.0)
+        
+        if "limit_tokens" in session_config:
+            result["limit_tokens"] = session_config["limit_tokens"]
+        if "limit_gpu_hours" in session_config:
+            result["limit_gpu_hours"] = session_config["limit_gpu_hours"]
+        
+        return result
+    except Exception as e:
+        click.echo(f"Warning: Failed to load session config from {config_path}: {e}", err=True)
+        return {"limit_cost_usd": 20.0}
+
+
+@click.command("opencode")
+@click.option(
+    "--model",
+    "model_name",
+    type=str,
+    default=None
+)
+@click.option(
+    "--force",
+    is_flag=True,
+    help="Prompt for API keys even if cached values exist."
+)
+@click.option(
+    "--url",
+    "override_url",
+    type=str,
+    default=None,
+)
+@click.option(
+    "--config",
+    "config_path",
+    type=click.Path(exists=True, path_type=Path),
+    default=None,
+    help="Path to TOML config file (default: opencode.toml or synth.toml in current directory)"
+)
+def opencode_cmd(
+    model_name: str | None = None,
+    force: bool = False,
+    override_url: str | None = None,
+    config_path: Path | None = None,
+) -> None:
+
+    while True:
+        bin_path = get_bin_path("opencode")
+        if bin_path:
+            break
+        if not install_bin(
+            "OpenCode",
+            [
+                "brew install opencode",
+                "bun add -g opencode-ai",
+                "curl -fsSL https://opencode.ai/install | bash",
+                "npm i -g opencode-ai",
+                "paru -S opencode"
+            ]
+        ):
+            print("Failed to find your installed OpenCode")
+            print("Please install from: https://opencode.ai")
+            return
+    print(f"Using OpenCode at {bin_path}")
+
+    if not verify_bin(bin_path):
+        print("Failed to verify OpenCode is runnable")
+        return
+
+    write_agents_md()
+
+    # Load session config from TOML
+    session_config = _load_session_config(config_path)
+    session_limit_cost = session_config.get("limit_cost_usd")
+    session_limit_tokens = session_config.get("limit_tokens")
+    session_limit_gpu_hours = session_config.get("limit_gpu_hours")
+    
+    # Get API key and base URL for session creation
+    synth_api_key = resolve_env_var("SYNTH_API_KEY", override_process_env=force)
+    if override_url:
+        base_url = override_url.rstrip("/")
+        if base_url.endswith("/api"):
+            base_url = base_url[:-4]
+    else:
+        base_url, _ = get_backend_from_env()
+    
+    # Create agent session with limits from config (default: $20 cost limit)
+    session_id: str | None = None
+    if session_limit_tokens or session_limit_cost or session_limit_gpu_hours:
+        try:
+            import asyncio
+            
+            # Create session - org_id will be fetched from backend /me endpoint
+            async def create_session():
+                from synth_ai.cli.local.session import AgentSessionClient
+                
+                client = AgentSessionClient(f"{base_url}/api", synth_api_key)
+                
+                limits = []
+                if session_limit_tokens:
+                    limits.append({
+                        "limit_type": "hard",
+                        "metric_type": "tokens",
+                        "limit_value": float(session_limit_tokens),
+                    })
+                if session_limit_cost:
+                    limits.append({
+                        "limit_type": "hard",
+                        "metric_type": "cost_usd",
+                        "limit_value": float(session_limit_cost),
+                    })
+                if session_limit_gpu_hours:
+                    limits.append({
+                        "limit_type": "hard",
+                        "metric_type": "gpu_hours",
+                        "limit_value": float(session_limit_gpu_hours),
+                    })
+                
+                # org_id will be automatically fetched from /api/v1/me endpoint
+                session = await client.create(
+                    org_id=None,  # Will be fetched from backend
+                    limits=limits,
+                    tracing_session_id=None,
+                    session_type="opencode_agent",
+                )
+                return session.session_id
+            
+            session_id = asyncio.run(create_session())
+            click.echo(f"✓ Created agent session: {session_id}")
+            if session_limit_tokens:
+                click.echo(f"  Token limit: {session_limit_tokens:,}")
+            if session_limit_cost:
+                click.echo(f"  Cost limit: ${session_limit_cost:.2f}")
+            if session_limit_gpu_hours:
+                click.echo(f"  GPU hours limit: {session_limit_gpu_hours:.2f}")
+        except Exception as e:
+            click.echo(f"Warning: Failed to create agent session: {e}", err=True)
+            click.echo("Continuing without session limits...", err=True)
+
+    if model_name is not None:
+        if model_name not in SYNTH_MODEL_NAMES:
+            raise ValueError(
+                f"model_name={model_name} is invalid. Valid values for model_name: {SYNTH_MODEL_NAMES}"
+            )
+        synth_api_key = resolve_env_var("SYNTH_API_KEY", override_process_env=force)
+        data = load_json_to_dict(AUTH_PATH)
+        good_entry = {
+            "type": "api",
+            "key": synth_api_key,
+        }
+        if data.get(SYNTH_PROVIDER_ID) != good_entry:
+            data[SYNTH_PROVIDER_ID] = good_entry
+        create_and_write_json(AUTH_PATH, data)
+        config = load_json_to_dict(CONFIG_PATH)
+        config.setdefault("$schema", "https://opencode.ai/config.json")
+        if override_url:
+            url = override_url
+            print("Using override URL:", url)
+        else:
+            url = BACKEND_URL_SYNTH_RESEARCH_BASE
+        provider_section = config.setdefault("provider", {})
+        synth_provider = provider_section.setdefault(SYNTH_PROVIDER_ID, {})
+        synth_provider["npm"] = "@ai-sdk/openai-compatible"
+        synth_provider.setdefault("name", "Synth")
+        models = synth_provider.setdefault("models", {})
+        models.setdefault(model_name, {})
+        options = synth_provider.setdefault("options", {})
+        options["baseURL"] = url
+        full_model_name = f"{SYNTH_PROVIDER_ID}/{model_name}"
+        config["model"] = full_model_name
+        create_and_write_json(CONFIG_PATH, config)
+
+    try:
+        subprocess.run([str(bin_path)], check=True)
+    except subprocess.CalledProcessError:
+        print("Failed to launch OpenCode")
+    finally:
+        # End session if created
+        if session_id is not None:
+            try:
+                # Assign to local variable to help type checker
+                final_session_id: str = session_id
+                import asyncio
+                
+                async def end_session():
+                    from synth_ai.cli.local.session import AgentSessionClient
+                    client = AgentSessionClient(f"{base_url}/api", synth_api_key)
+                    await client.end(final_session_id)
+                    click.echo(f"✓ Ended agent session: {final_session_id}")
+                
+                asyncio.run(end_session())
+            except Exception as e:
+                click.echo(f"Warning: Failed to end agent session: {e}", err=True)

synth_ai/cli/commands/__init__.py

@@ -0,0 +1,18 @@
+"""
+Structured CLI command implementations.
+
+Each subpackage under this namespace provides the core command entrypoints,
+validation helpers, and error types for a top-level CLI command (e.g. train,
+eval, deploy).
+"""
+
+from __future__ import annotations
+
+__all__ = [
+    "train",
+    "eval",
+    "filter",
+    "deploy",
+    "status",
+    "smoke",
+]

synth_ai/cli/commands/artifacts/__init__.py

@@ -0,0 +1,13 @@
+"""Artifacts CLI commands for managing models and prompts."""
+
+from __future__ import annotations
+
+import click
+
+
+def register(cli: click.Group) -> None:
+    """Register artifacts commands with the main CLI."""
+    from .core import artifacts_group
+    
+    cli.add_command(artifacts_group, name="artifacts")
+

synth_ai/cli/commands/artifacts/client.py

@@ -0,0 +1,119 @@
+"""HTTP client for artifacts API endpoints."""
+
+from __future__ import annotations
+
+import logging
+from typing import Any, Dict, List, Optional
+
+from synth_ai.core._utils.http import AsyncHttpClient
+
+logger = logging.getLogger(__name__)
+
+
+class ArtifactsClient:
+    """Client for artifacts API endpoints.
+    
+    Provides methods to interact with the Synth AI artifacts API, including
+    listing artifacts, retrieving model and prompt details, exporting models
+    to HuggingFace, and accessing prompt snapshots.
+    """
+    
+    def __init__(self, base_url: str, api_key: str, *, timeout: float = 30.0) -> None:
+        """Initialize the artifacts client.
+        
+        Args:
+            base_url: Base URL of the backend API (e.g., "https://api.usesynth.ai")
+            api_key: API key for authentication
+            timeout: Request timeout in seconds (default: 30.0)
+        """
+        self._base_url = base_url.rstrip("/")
+        self._api_key = api_key
+        self._timeout = timeout
+    
+    async def list_artifacts(
+        self,
+        *,
+        artifact_type: Optional[str] = None,
+        status: Optional[str] = None,
+        limit: int = 50,
+    ) -> Dict[str, Any]:
+        """List all artifacts (models and prompts)."""
+        params: Dict[str, Any] = {"limit": limit}
+        if artifact_type:
+            params["type"] = artifact_type
+        if status:
+            params["status"] = status
+        
+        async with AsyncHttpClient(self._base_url, self._api_key, timeout=self._timeout) as http:
+            return await http.get("/api/artifacts", params=params)
+    
+    async def get_model(self, model_id: str) -> Dict[str, Any]:
+        """Get detailed information about a model artifact."""
+        async with AsyncHttpClient(self._base_url, self._api_key, timeout=self._timeout) as http:
+            return await http.get(f"/api/artifacts/models/{model_id}")
+    
+    async def get_prompt(self, job_id: str) -> Dict[str, Any]:
+        """Get detailed information about a prompt optimization job."""
+        async with AsyncHttpClient(self._base_url, self._api_key, timeout=self._timeout) as http:
+            return await http.get(f"/api/artifacts/prompts/{job_id}")
+    
+    async def export_to_huggingface(
+        self,
+        *,
+        wasabi_key: str,
+        repo_id: str,
+        repo_type: str = "model",
+        artifact_kind: str = "lora",
+        base_model: Optional[str] = None,
+        visibility: str = "private",
+        tags: Optional[List[str]] = None,
+        bucket: Optional[str] = None,
+        folder_name: Optional[str] = None,
+    ) -> Dict[str, Any]:
+        """Export a model artifact to HuggingFace Hub."""
+        body: Dict[str, Any] = {
+            "repo_id": repo_id,
+            "repo_type": repo_type,
+            "artifact_kind": artifact_kind,
+            "key": wasabi_key,
+            "visibility": visibility,
+        }
+        if base_model:
+            body["base_model"] = base_model
+        if tags:
+            body["tags"] = tags
+        if bucket:
+            body["bucket"] = bucket
+        if folder_name:
+            body["folder_name"] = folder_name
+        
+        async with AsyncHttpClient(self._base_url, self._api_key, timeout=self._timeout) as http:
+            return await http.post_json("/api/learning/exports/hf", json=body)
+    
+    async def get_prompt_snapshot(
+        self,
+        job_id: str,
+        snapshot_id: str,
+    ) -> Dict[str, Any]:
+        """Get a specific prompt snapshot."""
+        async with AsyncHttpClient(self._base_url, self._api_key, timeout=self._timeout) as http:
+            return await http.get(f"/api/prompt-learning/online/jobs/{job_id}/snapshots/{snapshot_id}")
+    
+    async def list_prompt_snapshots(
+        self,
+        job_id: str,
+    ) -> List[Dict[str, Any]]:
+        """List all artifacts (snapshots) for a prompt job."""
+        async with AsyncHttpClient(self._base_url, self._api_key, timeout=self._timeout) as http:
+            artifacts = await http.get(f"/api/prompt-learning/online/jobs/{job_id}/artifacts")
+            if isinstance(artifacts, list):
+                return artifacts
+            if isinstance(artifacts, dict) and isinstance(artifacts.get("artifacts"), list):
+                return artifacts["artifacts"]
+            return []
+    
+    async def get_models_on_wasabi(self) -> Dict[str, Any]:
+        """Get models available on Wasabi."""
+        async with AsyncHttpClient(self._base_url, self._api_key, timeout=self._timeout) as http:
+            return await http.get("/api/learning/models/on-wasabi")
+