Busabase

Busabase is a workspace built specifically for AI agents to write into. Instead of agent output disappearing into chat logs or scratch files, it lands in typed databases, documents, and skills that persist across sessions. The core idea: agents propose changes, humans verify them, and only approved records become canonical data that other agents and apps can trust.

The review loop is the product's backbone. Every agent submission arrives as a structured change request with a field-level diff, a source context, and a full message history. Reviewers can approve, request revisions, or reject, and every decision stays attached to the record. Nothing enters your database until a human signs off.

Key capabilities:

• Change request review: inspect field-by-field diffs before any agent write touches your data
• Fact verification: check source context and downstream impact to catch hallucinations before they become business facts
• Audit trail: every submit, approval, and merge is logged with who, when, and why
• Canonical records: approved work becomes the source of truth with full commit history and relationship context
• Graph view: see how customers, content, assets, and business objects connect
• AirApps and dashboards: the same verified data renders directly into usable views without a separate build step
• Agent-first API: any agent that can call a REST API, including Claude Code, n8n, Zapier, or LangChain, can submit structured changes through OpenAPI, MCP, or Agent Skills

Busabase runs locally on PGlite, so your data never leaves your machine on the desktop version. Cloud adds collaboration, hosted APIs, and permission governance across teams. The open-source core is free to self-host; paid tiers add seats, storage, and API access.

It fits teams managing CRM enrichment, content pipelines, training datasets, finance records, or any workflow where AI proposes and humans need to verify before the output becomes institutional truth. The problem it solves is concrete: agent work that used to evaporate into sessions now accumulates into a database every subsequent agent can read and build on.