When to Use HatiData
HatiData is purpose-built for AI agents that need to store data, remember context, record reasoning, and govern access — all through a single Postgres connection. This guide helps you decide if it is the right fit.
Use HatiData When...
Your agents need persistent memory
Agents that need to remember across sessions — customer preferences, past interactions, learned facts — benefit from HatiData's hybrid SQL + vector memory system. Memories are stored in SQL tables, indexed with vector embeddings, and searchable by natural language or structured queries.
Example: A support agent remembers that a customer prefers email over phone, and retrieves this preference automatically during future interactions.
You need to audit agent reasoning
Regulated industries, financial services, and healthcare require proof of why an agent made a decision. HatiData's chain-of-thought ledger records every reasoning step in an immutable, cryptographically hash-chained format that is verifiable months later.
Example: A compliance officer replays an agent's decision to approve a loan, verifying every data point the agent considered.
You run multiple agents that share data
When multiple agents need coordinated access to the same data — with proper isolation, access control, and billing attribution — HatiData provides per-agent identity, ABAC policies, and row-level security out of the box.
Example: An analytics agent reads data, a reporting agent summarizes it, and an alerting agent monitors it — each with different scopes and billing.
Your agents run speculative queries
Agents that explore "what-if" scenarios — pricing changes, demand forecasts, risk models — use branch isolation to modify data without affecting production. Copy-on-write means branches are instant and memory-efficient.
Example: A research agent creates three branches to model different market scenarios, compares results, and discards the losing hypotheses.
You want semantic governance
Traditional rule-based alerting misses conceptual matches. Semantic triggers understand meaning — a trigger for "accessing personal data" fires on queries about SSNs, emails, or home addresses, even if the exact phrase never appears.
Example: A trigger alerts the security team whenever any agent's query semantically matches "exfiltrating customer data."
You need sub-10ms latency
Agents make hundreds of queries per reasoning chain. Second-per-query latency from legacy systems breaks the agent loop. HatiData's embedded query engine delivers sub-10ms analytical query latency with multi-tier caching.
Do Not Use HatiData When...
You only need a vector store
If your agents only need vector similarity search (no SQL, no reasoning traces, no governance), a dedicated vector database like Pinecone or Weaviate is a simpler choice. HatiData includes vector search as part of its hybrid system, but it is not optimized for vector-only workloads.
You need OLTP write patterns
HatiData is built for analytical (OLAP) workloads: aggregations, scans, and joins. It is not designed for high-frequency single-row transactional writes (thousands of INSERTs per second). Use PostgreSQL or another OLTP database for transactional workloads, and connect HatiData for the analytical layer.
Your agents do not need identity or audit
If your agents connect anonymously and you do not need per-agent billing, access control, or audit trails, a simpler setup (SQLite or a managed database) may be sufficient. HatiData's value comes from the governance layer.
You need real-time streaming
HatiData processes batch and interactive queries. It does not support real-time streaming ingestion (like Kafka Streams or Apache Flink). For streaming pipelines, use a streaming engine and sink results into HatiData for agent access.
What About...
| Alternative | When to Choose It | When HatiData Is Better |
|---|---|---|
| PostgreSQL | OLTP workloads, simple agent setups | Per-agent identity, memory, CoT, triggers, sub-10ms analytics |
| Legacy Cloud Warehouses | Existing large-scale BI infrastructure | Agent-native features, sub-10ms latency, in-VPC deployment |
| Pinecone / Weaviate | Vector-only search, no SQL needed | Hybrid SQL + vector, governance, reasoning traces |
| SQLite / embedded databases | Single-agent prototypes, no governance | Multi-agent, audit, policies, memory, triggers |
| Redis | Caching, session state | Persistent analytical queries, complex SQL, audit |
| LangChain Memory | Simple conversation memory | Long-term cross-session memory, SQL queries, vector hybrid |
Decision Flowchart
- Do your agents need SQL query access? → If no, consider a vector store or key-value store.
- Do you need per-agent identity and billing? → If no, consider a simpler embedded database.
- Do you need audit trails for agent decisions? → If no, consider a simpler database.
- Do your agents need persistent memory? → If no, consider PostgreSQL.
- If you answered yes to 2 or more → HatiData is likely the right fit.
Next Steps
- Quickstart — Try it in under 5 minutes
- What is ANDI? — Understand the category
- Architecture in 60 Seconds — How it works
- Migration Guides — Coming from another system