haiku.rag 0.11.4__tar.gz → 0.12.0__tar.gz

{haiku_rag-0.11.4 → haiku_rag-0.12.0}/.gitignore

@@ -21,3 +21,7 @@ tests/data/
 TODO.md
 PLAN.md
 DEVNOTES.md
+
+# mcp registry
+.mcpregistry_github_token
+.mcpregistry_registry_token

haiku_rag-0.12.0/.python-version

@@ -0,0 +1 @@
+3.13

{haiku_rag-0.11.4 → haiku_rag-0.12.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: haiku.rag
-Version: 0.11.4
+Version: 0.12.0
 Summary: Agentic Retrieval Augmented Generation (RAG) with LanceDB
 Author-email: Yiorgis Gozadinos <ggozadinos@gmail.com>
 License: MIT
@@ -18,18 +18,20 @@ Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Classifier: Typing :: Typed
 Requires-Python: >=3.12
-Requires-Dist: docling>=2.52.0
-Requires-Dist: fastmcp>=2.12.3
+Requires-Dist: docling>=2.56.1
+Requires-Dist: fastmcp>=2.12.4
 Requires-Dist: httpx>=0.28.1
-Requires-Dist: lancedb>=0.25.0
-Requires-Dist: pydantic-ai>=1.0.8
-Requires-Dist: pydantic-graph>=1.0.8
-Requires-Dist: pydantic>=2.11.9
+Requires-Dist: lancedb>=0.25.2
+Requires-Dist: pydantic-ai>=1.0.18
+Requires-Dist: pydantic-graph>=1.0.18
+Requires-Dist: pydantic>=2.12.1
 Requires-Dist: python-dotenv>=1.1.1
-Requires-Dist: rich>=14.1.0
-Requires-Dist: tiktoken>=0.11.0
-Requires-Dist: typer>=0.16.1
+Requires-Dist: rich>=14.2.0
+Requires-Dist: tiktoken>=0.12.0
+Requires-Dist: typer>=0.19.2
 Requires-Dist: watchfiles>=1.1.0
+Provides-Extra: a2a
+Requires-Dist: fasta2a>=0.1.0; extra == 'a2a'
 Provides-Extra: mxbai
 Requires-Dist: mxbai-rerank>=0.1.6; extra == 'mxbai'
 Provides-Extra: voyageai
@@ -56,6 +58,7 @@ Retrieval-Augmented Generation (RAG) library built on LanceDB.
 - **File monitoring**: Auto-index files when run as server
 - **40+ file formats**: PDF, DOCX, HTML, Markdown, code files, URLs
 - **MCP server**: Expose as tools for AI assistants
+- **A2A agent**: Conversational agent with context and multi-turn dialogue
 - **CLI & Python API**: Use from command line or Python
 
 ## Quick Start
@@ -181,6 +184,24 @@ haiku-rag serve --stdio
 
 Provides tools for document management and search directly in your AI assistant.
 
+## A2A Agent
+
+Run as a conversational agent with the Agent-to-Agent protocol:
+
+```bash
+# Start the A2A server
+haiku-rag serve --a2a
+
+# Connect with the interactive client (in another terminal)
+haiku-rag a2aclient
+```
+
+The A2A agent provides:
+- Multi-turn dialogue with context
+- Intelligent multi-search for complex questions
+- Source citations with titles and URIs
+- Full document retrieval on request
+
 ## Documentation
 
 Full documentation at: https://ggozad.github.io/haiku.rag/

{haiku_rag-0.11.4 → haiku_rag-0.12.0}/README.md

@@ -18,6 +18,7 @@ Retrieval-Augmented Generation (RAG) library built on LanceDB.
 - **File monitoring**: Auto-index files when run as server
 - **40+ file formats**: PDF, DOCX, HTML, Markdown, code files, URLs
 - **MCP server**: Expose as tools for AI assistants
+- **A2A agent**: Conversational agent with context and multi-turn dialogue
 - **CLI & Python API**: Use from command line or Python
 
 ## Quick Start
@@ -143,6 +144,24 @@ haiku-rag serve --stdio
 
 Provides tools for document management and search directly in your AI assistant.
 
+## A2A Agent
+
+Run as a conversational agent with the Agent-to-Agent protocol:
+
+```bash
+# Start the A2A server
+haiku-rag serve --a2a
+
+# Connect with the interactive client (in another terminal)
+haiku-rag a2aclient
+```
+
+The A2A agent provides:
+- Multi-turn dialogue with context
+- Intelligent multi-search for complex questions
+- Source citations with titles and URIs
+- Full document retrieval on request
+
 ## Documentation
 
 Full documentation at: https://ggozad.github.io/haiku.rag/

{haiku_rag-0.11.4 → haiku_rag-0.12.0}/mkdocs.yml

@@ -64,6 +64,7 @@ nav:
       - Agents: agents.md
       - Python: python.md
       - MCP: mcp.md
+      - A2A: a2a.md
       - Benchmarks: benchmarks.md
 markdown_extensions:
   - admonition

{haiku_rag-0.11.4 → haiku_rag-0.12.0}/pyproject.toml

@@ -2,7 +2,7 @@
 
 name = "haiku.rag"
 description = "Agentic Retrieval Augmented Generation (RAG) with LanceDB"
-version = "0.11.4"
+version = "0.12.0"
 authors = [{ name = "Yiorgis Gozadinos", email = "ggozadinos@gmail.com" }]
 license = { text = "MIT" }
 readme = { file = "README.md", content-type = "text/markdown" }
@@ -23,23 +23,24 @@ classifiers = [
 ]
 
 dependencies = [
-    "docling>=2.52.0",
-    "fastmcp>=2.12.3",
+    "docling>=2.56.1",
+    "fastmcp>=2.12.4",
     "httpx>=0.28.1",
-    "lancedb>=0.25.0",
-    "pydantic>=2.11.9",
-    "pydantic-ai>=1.0.8",
-    "pydantic-graph>=1.0.8",
+    "lancedb>=0.25.2",
+    "pydantic>=2.12.1",
+    "pydantic-ai>=1.0.18",
+    "pydantic-graph>=1.0.18",
     "python-dotenv>=1.1.1",
-    "rich>=14.1.0",
-    "tiktoken>=0.11.0",
-    "typer>=0.16.1",
+    "rich>=14.2.0",
+    "tiktoken>=0.12.0",
+    "typer>=0.19.2",
     "watchfiles>=1.1.0",
 ]
 
 [project.optional-dependencies]
 voyageai = ["voyageai>=0.3.5"]
 mxbai = ["mxbai-rerank>=0.1.6"]
+a2a = ["fasta2a>=0.1.0"]
 
 [project.scripts]
 haiku-rag = "haiku.rag.cli:cli"
@@ -49,7 +50,7 @@ requires = ["hatchling"]
 build-backend = "hatchling.build"
 
 [tool.hatch.build]
-exclude = ["/docs", "/tests", "/.github"]
+exclude = ["/docs", "/examples", "/tests", "/.github"]
 
 [tool.hatch.build.targets.wheel]
 packages = ["src/haiku"]

haiku_rag-0.12.0/server.json

@@ -0,0 +1,253 @@
+{
+  "$schema": "https://static.modelcontextprotocol.io/schemas/2025-09-29/server.schema.json",
+  "name": "io.github.ggozad/haiku-rag",
+  "version": "{{VERSION}}",
+  "description": "Agentic Retrieval Augmented Generation (RAG) with LanceDB",
+  "repository": {
+    "url": "https://github.com/ggozad/haiku.rag",
+    "source": "github"
+  },
+  "homepage": "https://github.com/ggozad/haiku.rag",
+  "license": "MIT",
+  "keywords": ["rag", "lancedb", "vector-database", "embeddings", "search", "qa", "research"],
+  "vendor": {
+    "name": "Yiorgis Gozadinos",
+    "url": "https://github.com/ggozad"
+  },
+  "deployment": {
+    "packages": [
+      {
+        "type": "pypi",
+        "package": "haiku.rag",
+        "command": {
+          "linux-x86_64": {
+            "shell": "uvx",
+            "args": ["haiku.rag", "serve", "--stdio"]
+          },
+          "darwin-arm64": {
+            "shell": "uvx",
+            "args": ["haiku.rag", "serve", "--stdio"]
+          },
+          "darwin-x86_64": {
+            "shell": "uvx",
+            "args": ["haiku.rag", "serve", "--stdio"]
+          },
+          "win32-x86_64": {
+            "shell": "uvx.exe",
+            "args": ["haiku.rag", "serve", "--stdio"]
+          }
+        },
+        "environmentVariables": [
+          {
+            "name": "ENV",
+            "description": "Runtime environment (production or development)",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "DEFAULT_DATA_DIR",
+            "description": "Default directory for LanceDB data and assets",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "MONITOR_DIRECTORIES",
+            "description": "Comma-separated paths to watch for file changes in server mode",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "LANCEDB_URI",
+            "description": "LanceDB connection URI (use db:// for cloud or a filesystem path)",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "LANCEDB_REGION",
+            "description": "LanceDB cloud region (if using cloud)",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "LANCEDB_API_KEY",
+            "description": "LanceDB API key (required for LanceDB Cloud)",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": true
+          },
+          {
+            "name": "EMBEDDINGS_PROVIDER",
+            "description": "Embeddings provider (e.g. ollama, openai, voyageai)",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "EMBEDDINGS_MODEL",
+            "description": "Embeddings model name (provider-specific)",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "EMBEDDINGS_VECTOR_DIM",
+            "description": "Embedding vector dimension (must match model)",
+            "format": "number",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "QA_PROVIDER",
+            "description": "Question answering provider (e.g. ollama, openai, anthropic)",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "QA_MODEL",
+            "description": "Question answering model name (provider-specific)",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "RESEARCH_PROVIDER",
+            "description": "Research provider for multi-agent research (e.g. ollama, openai, anthropic)",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "RESEARCH_MODEL",
+            "description": "Research model name for multi-agent research (provider-specific)",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "RERANK_PROVIDER",
+            "description": "Rerank provider (e.g. mixedbread, cohere)",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "RERANK_MODEL",
+            "description": "Rerank model name (provider-specific)",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "CHUNK_SIZE",
+            "description": "Chunk size for splitting documents (characters)",
+            "format": "number",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "CONTEXT_CHUNK_RADIUS",
+            "description": "Number of adjacent chunks to include around search hits",
+            "format": "number",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "OLLAMA_BASE_URL",
+            "description": "Base URL for Ollama server",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "VLLM_EMBEDDINGS_BASE_URL",
+            "description": "Base URL for vLLM embeddings endpoint",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "VLLM_RERANK_BASE_URL",
+            "description": "Base URL for vLLM rerank endpoint",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "VLLM_QA_BASE_URL",
+            "description": "Base URL for vLLM QA endpoint",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "VLLM_RESEARCH_BASE_URL",
+            "description": "Base URL for vLLM research endpoint",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "MARKDOWN_PREPROCESSOR",
+            "description": "Dotted path or file path to a callable that preprocesses markdown content before chunking",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "DISABLE_DB_AUTOCREATE",
+            "description": "If true, refuse to auto-create a new LanceDB database or tables",
+            "format": "boolean",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "VACUUM_RETENTION_SECONDS",
+            "description": "Vacuum retention threshold in seconds (default: 60)",
+            "format": "number",
+            "isRequired": false,
+            "isSecret": false
+          },
+          {
+            "name": "OPENAI_API_KEY",
+            "description": "OpenAI API key (if using OpenAI for embeddings or QA)",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": true
+          },
+          {
+            "name": "VOYAGE_API_KEY",
+            "description": "VoyageAI API key (if using VoyageAI for embeddings)",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": true
+          },
+          {
+            "name": "ANTHROPIC_API_KEY",
+            "description": "Anthropic API key (if using Anthropic for QA)",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": true
+          },
+          {
+            "name": "COHERE_API_KEY",
+            "description": "Cohere API key (if using Cohere for reranking)",
+            "format": "string",
+            "isRequired": false,
+            "isSecret": true
+          }
+        ]
+      }
+    ]
+  },
+  "transports": [
+    {
+      "type": "stdio"
+    }
+  ]
+}

{haiku_rag-0.11.4 → haiku_rag-0.12.0}/src/evaluations/benchmark.py

@@ -212,8 +212,6 @@ async def run_qa_benchmark(
                 return await qa.answer(question)
 
             for case in evaluation_dataset.cases:
-                progress.console.print(f"\n[bold]Evaluating case:[/bold] {case.name}")
-
                 single_case_dataset = EvalDataset[str, str, dict[str, str]](
                     cases=[case],
                     evaluators=evaluation_dataset.evaluators,
@@ -232,32 +230,24 @@ async def run_qa_benchmark(
                     result_case = report.cases[0]
 
                     equivalence = result_case.assertions.get("answer_equivalent")
-                    progress.console.print(f"Question: {result_case.inputs}")
-                    progress.console.print(f"Expected: {result_case.expected_output}")
-                    progress.console.print(f"Generated: {result_case.output}")
                     if equivalence is not None:
-                        progress.console.print(
-                            f"Equivalent: {equivalence.value}"
-                            + (f" — {equivalence.reason}" if equivalence.reason else "")
-                        )
                         if equivalence.value:
                             passing_cases += 1
 
-                    progress.console.print("")
-
                 if report.failures:
                     failures.extend(report.failures)
                     failure = report.failures[0]
                     progress.console.print(
                         "[red]Failure encountered during case evaluation:[/red]"
                     )
-                    progress.console.print(f"Question: {failure.inputs}")
                     progress.console.print(f"Error: {failure.error_message}")
                     progress.console.print("")
 
-                progress.console.print(
-                    f"[green]Accuracy: {(passing_cases / total_processed):.4f} "
-                    f"{passing_cases}/{total_processed}[/green]"
+                progress.update(
+                    qa_task,
+                    description="[yellow]Evaluating QA cases...[/yellow] "
+                    f"[green]Accuracy: {(passing_cases / total_processed):.2f} "
+                    f"{passing_cases}/{total_processed}[/green]",
                 )
                 progress.advance(qa_task)
 

haiku_rag-0.12.0/src/haiku/rag/a2a/__init__.py

@@ -0,0 +1,176 @@
+import logging
+from contextlib import asynccontextmanager
+from pathlib import Path
+
+import logfire
+from pydantic_ai import Agent, RunContext
+
+from haiku.rag.config import Config
+from haiku.rag.graph.common import get_model
+
+from .context import load_message_history, save_message_history
+from .models import AgentDependencies, SearchResult
+from .prompts import A2A_SYSTEM_PROMPT
+from .skills import extract_question_from_task, get_agent_skills
+from .storage import LRUMemoryStorage
+from .worker import ConversationalWorker
+
+try:
+    from fasta2a import FastA2A  # type: ignore
+    from fasta2a.broker import InMemoryBroker  # type: ignore
+    from fasta2a.storage import InMemoryStorage  # type: ignore
+except ImportError as e:
+    raise ImportError(
+        "A2A support requires the 'a2a' extra. "
+        "Install with: uv pip install 'haiku.rag[a2a]'"
+    ) from e
+
+logfire.configure(send_to_logfire="if-token-present", service_name="a2a")
+logfire.instrument_pydantic_ai()
+
+logger = logging.getLogger(__name__)
+
+__all__ = [
+    "create_a2a_app",
+    "load_message_history",
+    "save_message_history",
+    "extract_question_from_task",
+    "get_agent_skills",
+    "LRUMemoryStorage",
+]
+
+
+def create_a2a_app(
+    db_path: Path,
+    security_schemes: dict | None = None,
+    security: list[dict[str, list[str]]] | None = None,
+):
+    """Create an A2A app for the conversational QA agent.
+
+    Args:
+        db_path: Path to the LanceDB database
+        security_schemes: Optional security scheme definitions for the AgentCard
+        security: Optional security requirements for the AgentCard
+
+    Returns:
+        A FastA2A ASGI application
+    """
+    base_storage = InMemoryStorage()
+    storage = LRUMemoryStorage(
+        storage=base_storage, max_contexts=Config.A2A_MAX_CONTEXTS
+    )
+    broker = InMemoryBroker()
+
+    # Create the agent with native search tool
+    model = get_model(Config.QA_PROVIDER, Config.QA_MODEL)
+    agent = Agent(
+        model=model,
+        deps_type=AgentDependencies,
+        system_prompt=A2A_SYSTEM_PROMPT,
+        retries=3,
+    )
+
+    @agent.tool
+    async def search_documents(
+        ctx: RunContext[AgentDependencies],
+        query: str,
+        limit: int = 3,
+    ) -> list[SearchResult]:
+        """Search the knowledge base for relevant documents.
+
+        Returns chunks of text with their relevance scores and document URIs.
+        Use get_full_document if you need to see the complete document content.
+        """
+        search_results = await ctx.deps.client.search(query, limit=limit)
+        expanded_results = await ctx.deps.client.expand_context(search_results)
+
+        return [
+            SearchResult(
+                content=chunk.content,
+                score=score,
+                document_title=chunk.document_title,
+                document_uri=(chunk.document_uri or ""),
+            )
+            for chunk, score in expanded_results
+        ]
+
+    @agent.tool
+    async def get_full_document(
+        ctx: RunContext[AgentDependencies],
+        document_uri: str,
+    ) -> str:
+        """Retrieve the complete content of a document by its URI.
+
+        Use this when you need more context than what's in a search result chunk.
+        The document_uri comes from search_documents results.
+        """
+        document = await ctx.deps.client.get_document_by_uri(document_uri)
+        if document is None:
+            return f"Document not found: {document_uri}"
+
+        return document.content
+
+    worker = ConversationalWorker(
+        storage=storage,
+        broker=broker,
+        db_path=db_path,
+        agent=agent,  # type: ignore
+    )
+
+    # Create FastA2A app with custom worker lifecycle
+    @asynccontextmanager
+    async def lifespan(app):
+        logger.info(f"Started A2A server (max contexts: {Config.A2A_MAX_CONTEXTS})")
+        async with app.task_manager:
+            async with worker.run():
+                yield
+
+    app = FastA2A(
+        storage=storage,
+        broker=broker,
+        name="haiku-rag",
+        description="Conversational question answering agent powered by haiku.rag RAG system",
+        skills=get_agent_skills(),
+        lifespan=lifespan,
+    )
+
+    # Add security configuration if provided
+    if security_schemes or security:
+        # Monkey-patch the agent card endpoint to include security
+        async def _agent_card_endpoint_with_security(request):
+            from fasta2a.schema import AgentCapabilities, AgentCard, agent_card_ta
+            from starlette.responses import Response
+
+            if app._agent_card_json_schema is None:
+                agent_card = AgentCard(
+                    name=app.name,
+                    description=app.description
+                    or "An AI agent exposed as an A2A agent.",
+                    url=app.url,
+                    version=app.version,
+                    protocol_version="0.3.0",
+                    skills=app.skills,
+                    default_input_modes=app.default_input_modes,
+                    default_output_modes=app.default_output_modes,
+                    capabilities=AgentCapabilities(
+                        streaming=False,
+                        push_notifications=False,
+                        state_transition_history=False,
+                    ),
+                )
+                if app.provider is not None:
+                    agent_card["provider"] = app.provider
+                if security_schemes:
+                    agent_card["security_schemes"] = security_schemes
+                if security:
+                    agent_card["security"] = security
+                app._agent_card_json_schema = agent_card_ta.dump_json(
+                    agent_card, by_alias=True
+                )
+            return Response(
+                content=app._agent_card_json_schema, media_type="application/json"
+            )
+
+        app._agent_card_endpoint = _agent_card_endpoint_with_security
+
+    return app