Structured records
Title, learning, acceptance criteria, category, 1 to 5 tags, priority 1 to 5 and an automatic version on every record.
A shared memory for your AI agents: every learning is stored once, as a structured record, and every agent checks it before the next task.
SynthBrain is a small knowledge store for LLM agents. When a coding agent finds out that a host rejects a certain database collation, that insight usually dies with the session, and the next agent finds it out again. SynthBrain keeps it: the agent writes one record with a title, the learning, acceptance criteria, a category, tags and a priority, and every later agent searches for it before it starts.
Agents talk to SynthBrain through an XML API with search, batch reads and bulk writes of up to 50 records, all or nothing. Each agent gets its own API key with exactly the scopes it needs (read, write, update or full), and a public {/synthbrain/llms.txt} file explains the whole API, so connecting a new agent takes one key and one link.
The owner reviews, edits and exports records in a browser admin, manages categories and keys, and sees on a dashboard how the store grows. It runs on ordinary shared hosting with PHP and MySQL and no extra services, which I think matters more than any feature for a tool agents call at the start of nearly every task. SynthBrain is available on request: we set it up for your team or on your own hosting.
These are the real numbers from our own SynthBrain, the one our coding and automation agents read before they start work and write to after they learn something.
Updated 22 November 2026, 10:25 UTC, refreshed once a day · 4 more categories are set up but still empty
Title, learning, acceptance criteria, category, 1 to 5 tags, priority 1 to 5 and an automatic version on every record.
Keyword, category, tag, priority and date filters; results sorted by priority so the rules agents must follow come first.
An XSD schema rejects malformed input precisely, and agents can validate their own requests before they send them.
Create or update up to 50 records in one call, all or nothing, and read up to 50 by ID.
One key per agent with read, write, update or full scope; only full may delete, and a key is revoked in one click.
A second record with the same title is refused, so agents update a learning instead of duplicating it.
Records table with filters, record editor, categories, keys, XML export and a dashboard of growth and agent use.
Admin login with lockout, an IP block after repeated invalid keys, and keys that never appear in URLs.
Self-hosted PHP and MySQL, an XML API for agents and a browser admin for the owner
Available on request. Tell us how many agents you run and where you host, and we set SynthBrain up for you. The API spec for agents is public at /synthbrain/llms.txt.
SYNTHETIXMIND builds tools like this, or similar ones, custom-made for our customers: calculators, planners, configurators and quote tools in your brand, on your website, built around your own rules and data.
Ideas, feedback or a custom version for your company? Talk to us.
Contact Us →