Back to browse
GitHub Repository

Jaybase is a append-only fact store for AI agents doing critical business workflows

1 starsGo

Jaybase: Append-only fact store for AI doing business workflows

by kvisner·Jul 21, 2026·2 points·0 comments

AI Analysis

●●●BangerBig BrainZero to One

Merkle DAG storage prevents runaway agents from corrupting data without leaving an audit trail.

Strengths
  • Append-only architecture ensures every agent decision remains visible and correctable forever.
  • Idempotency keys and stale-root checks prevent duplicate writes and race conditions automatically.
  • Flexible JSON schema allows evolving fact structures without breaking historical data integrity.
Weaknesses
  • Single-tenant design limits applicability for SaaS providers needing multi-tenant isolation.
  • Go-based server requires manual deployment compared to managed database alternatives.
Category
Target Audience

Developers building AI agents for critical business workflows

Similar To

EventStoreDB · Datomic · Hyperledger Fabric

Post Description

Hey HN,

I build custom AI agents for businesses, and one of the things I've noticed is that everyone wants to use AI to do real, critical work, but it's hard to trust it because sometimes agents can go a little off the rails. No one wants their agent deleting all their customer data or screwing up their books. So they tend to limit its use to searching, summarizing, and as a writing assistant.

The problem is that most ways we store data, from notes & spreadsheets to databases, were designed for humans who work at and make mistakes at human speed, or for deterministic software systems. AI agents work non-deterministically at machine speed, which is a new challenge.

That's why I made Jaybase. Jaybase is an append-only fact store for agentic workflows. It's based on a Merkle Directed Acyclic Graph, which is the same technology behind git and the blockchain, so every write is auditable, replayable, and reversible. If an agent goes off the rails and corrupts a bunch of data, it's easy to recover and easy to audit what went wrong. It also doesn't have a fixed schema, so as you add agents or your agents get more skills, it's natural to add more kinds of data.

Happy to dive into any questions in the comments and would love the community’s feedback. Thanks!

Similar Projects

AI/ML●●●Banger

Mengram – AI agent memory with facts, events, and evolving workflows

Procedures that learn from failures—not just facts; agents improve through experience loops.

Zero to OneBig Brain
mengram-ai
214mo ago