HatiData Agent SDK

RAM for Agents — Python SDK for AI agents to query HatiData's in-VPC data warehouse with sub-10ms latency via Postgres wire protocol.

Installation

pip install hatidata-agent

# With async support
pip install hatidata-agent[async]

# With LangChain support
pip install hatidata-agent[langchain]

# With MCP server
pip install hatidata-agent[mcp]

# Everything
pip install hatidata-agent[all]

Quick Start

from hatidata_agent import HatiDataAgent

agent = HatiDataAgent(
    host="proxy.internal",
    port=5439,
    agent_id="my-agent",
    framework="langchain",
)

# Simple query
rows = agent.query("SELECT * FROM customers WHERE status = 'active' LIMIT 10")
print(rows)  # [{"id": 1, "name": "Acme Corp", ...}, ...]

Features

• Sub-10ms query latency -- In-VPC execution, no data leaves your network
• Postgres wire protocol -- Works with any Postgres client library
• Reasoning chain tracking -- Multi-step audit trails for agent workflows
• RAG context retrieval -- Full-text and vector similarity search
• LangChain integration -- Drop-in SQLDatabase replacement for LangChain agents
• MCP server -- Expose HatiData as tools for Claude and other MCP-compatible agents
• Agent identification -- Per-agent billing, priority scheduling, and audit via Postgres startup parameters
• Snowflake SQL compatible -- Bring existing queries without rewrites

Reasoning Chain Tracking

Track multi-step reasoning chains for audit and debugging:

with agent.reasoning_chain("req-001") as chain:
    # Step 0: Discover tables
    tables = chain.query("SELECT table_name FROM information_schema.tables")

    # Step 1: Get relevant data
    customers = chain.query("SELECT * FROM customers WHERE tier = 'enterprise'", step=1)

    # Step 2: Aggregate
    revenue = chain.query("SELECT SUM(revenue) FROM orders WHERE customer_id IN (...)", step=2)

RAG Context Retrieval

# Full-text search
context = agent.get_context("customers", "enterprise accounts in US", top_k=5)

# Vector similarity search
context = agent.get_rag_context("docs", "embedding", query_vector, top_k=10)

LangChain Integration

from hatidata_agent.langchain import HatiDataSQLDatabase
from langchain.agents import create_sql_agent

db = HatiDataSQLDatabase(
    host="proxy.internal",
    port=5439,
    agent_id="sql-agent-1",
)

agent = create_sql_agent(llm=llm, db=db, verbose=True)
result = agent.run("How many enterprise customers do we have?")

MCP Server

Run as an MCP server for Claude and other MCP-compatible agents:

# Start the MCP server
hatidata-mcp-server --host proxy.internal --port 5439

# Or add to Claude Code's MCP config:
# ~/.claude/mcp.json
{
  "mcpServers": {
    "hatidata": {
      "command": "hatidata-mcp-server",
      "args": ["--host", "proxy.internal", "--port", "5439"]
    }
  }
}

MCP Tools

Agent Identification

The SDK automatically identifies agents via Postgres startup parameters:

These enable per-agent billing, priority scheduling, audit trails, and the Agent Tax Report showing savings vs legacy cloud warehouses.

Documentation

Full documentation is available at docs.hatiosai.com/hatidata.

License

Apache License 2.0. Copyright (c) Marviy Pte Ltd. See LICENSE for details.