crewai-hydradb

HydraDB as external memory for CrewAI agents. Your crews recall relevant history on each task and persist their outputs into HydraDB, so memory survives across runs.

Install

pip install crewai-hydradb   # once published; for local dev: pip install -e .

Prerequisites

• Python >= 3.10
• A HydraDB account: API key + tenant ID (hydradb.com)
• crewai >= 1.0, < 1.11

CrewAI version compatibility

This integration implements CrewAI's Storage interface (crewai.memory.storage.interface.Storage) wired via ExternalMemory, which exists in CrewAI 1.0-1.10 (verified present in 1.5, removed in 1.11). CrewAI 1.11+ restructured memory around a new embedding-based StorageBackend protocol that expects the backend to store and search raw embedding vectors - a poor fit for HydraDB's text-query recall. On 1.11+, use HydraDB over MCP instead (hydra-db/hydradb-mcp), which exposes HydraDB's own retrieval without a vector-store contract.

Quick start

from crewai import Crew
from crewai.memory.external.external_memory import ExternalMemory
from hydradb_crewai import HydraDBClient, HydraDBStorage

client = HydraDBClient(api_key="sk_live_...", tenant_id="your-tenant")

crew = Crew(
    agents=[...],
    tasks=[...],
    external_memory=ExternalMemory(storage=HydraDBStorage(client)),
)
crew.kickoff()

Notes

• Public API only. save writes via the public knowledge path (app_knowledge → /context/ingest); search uses POST /query. The memory-family write route is not publicly exposed, so outputs are stored as knowledge (default kind="knowledge").

How it works

CrewAI's ExternalMemory delegates to a Storage implementation:

This mirrors the pattern used by Mem0, Honcho, and Hindsight.

Test

pip install -e ".[test]"
pytest

License

Apache-2.0 - Copyright (c) 2026 HydraDB