HatiData Documentation
Agent-Native Data Infrastructure for AI agents. Sub-10ms SQL, persistent memory, verifiable reasoning, semantic triggers — all through a single Postgres connection.
What is HatiData?
HatiData is Agent-Native Data Infrastructure (ANDI) — the data layer purpose-built for AI agents. It speaks the Postgres wire protocol, transpiles Snowflake-compatible SQL for execution, and delivers sub-10ms query latency. Beyond SQL, HatiData gives agents persistent memory, semantic triggers, chain-of-thought logging, and branch isolation out of the box.
Read the full ANDI definition →
Five Pillars of ANDI
| Pillar | What It Does | Learn More |
|---|---|---|
| Agent Identity | Per-agent keys, scopes, billing, and audit attribution | Identity Model |
| Persistent Memory | SQL + vector hybrid search across sessions | Memory |
| Chain-of-Thought | Immutable cryptographically hash-chained reasoning traces | CoT Ledger |
| Semantic Triggers | Concept-based rules with webhook, notify, and review actions | Triggers |
| Branch Isolation | Copy-on-write speculative queries with merge strategies | Branching |
How It Works
Every SQL query flows through a multi-stage pipeline that handles access control, SQL transpilation, execution, column masking, metering, and audit automatically. Security and compliance are built into every query path — not bolted on after.
For the full pipeline, see Architecture in 60 Seconds.
Three Deployment Tiers
| Tier | Price | Data Location | Best For |
|---|---|---|---|
| Local | Free | Your machine | Development, prototyping, CI |
| Cloud | $29/month | HatiData-managed | Small teams, startups |
| Enterprise | Custom | Your VPC | Regulated industries, production |
The same SQL, SDKs, and agent code work across all tiers. Only the connection string changes.
Key Differentiators
- Agent-native — Agents identify themselves via startup parameters. Per-agent billing, scheduling, and audit happen automatically.
- Sub-10ms latency — Embedded columnar engine with three-tier caching (in-memory, high-performance local SSD, object storage).
- Snowflake SQL compatible — The transpiler handles function mapping (NVL, IFF, DATEDIFF), type mapping (VARIANT, TIMESTAMP_NTZ), and construct rewriting (FLATTEN, OBJECT_CONSTRUCT).
- Secure by default — Access control, column masking, metering, and audit built into every query path.
- Local-first — Start with
hati initand query immediately. No cloud account needed.
Integrations
HatiData works with every major AI agent framework and any Postgres-compatible tool:
- MCP — Claude, Cursor, and MCP-compatible agents
- LangChain — Memory, VectorStore, Toolkit
- CrewAI — Multi-agent memory and tools
- AutoGen — Group memory and coordination
- dbt — Transform agent data with dbt models
- BI Tools — Metabase, Grafana, Tableau, DBeaver
Open Source
The CLI, SDKs, and integrations are open source under the Apache-2.0 license:
- GitHub: HatiOS-AI/HatiData-SDKs — Python SDK, TypeScript SDK, dbt adapter, LangChain, CrewAI, MCP configs, and examples
V2: Governed Runtime
HatiData V2 adds a Governed Runtime — lifecycle management for agent tasks, with full lineage, explainability, and human-in-the-loop checkpoints.
| V2 Feature | What It Does | Learn More |
|---|---|---|
| Tasks & Attempts | Intent vs execution separation, leases, recovery | Tasks & Attempts |
| Lineage & Explainability | Trace any artifact to its prompt, model, and cost | Lineage |
| Branching & Isolation | 4 visibility modes, no-leak guarantee | Branching |
| Human-in-the-Loop | Gate predicates, review queues, recovery workflows | HITL |
| Optimization & Replay | Reward signals, bandit routing, counterfactual replay | Optimization |
Migrating from V1? See the V1 to V2 Transition Guide.
Next Steps
- Quickstart — Install and run your first query in under 5 minutes
- What is ANDI? — Understand the category
- Governed Runtime (V2) — The new agent lifecycle layer
- Python SDK — Agent-aware access for Python
- Enterprise — In-VPC architecture with PrivateLink