Data and knowledge
Three platform-managed nodes give agents persistent state. In all three, the node's description is read by the model: it says what the node contains, and the agent reads it before searching.
The same database can give full access to a developer agent and read-only access to an analysis agent.
Procedures written as skills are loaded only when needed and shared across agents.
MySQL database
A database created and maintained by the platform, queryable by agents and code and viewable from the Databases section (schema, queries, data).
| Port | Privileges | Typical use |
|---|---|---|
| admin (Full access) | Structure and data | Developer agent or code that creates the schema |
| rw (Read & write) | Read, insert, update, delete data | Agents and apps in operation |
| ro (Read only) | Read only, with an optional table filter | Analysis and reporting |
- Every port offers
schema, which reads tables, columns and keys in real time, andquery, which runs a single statement per call with parameters. - A query returns at most 100 rows in the reply; a larger result is attached in full as a CSV file.
- A schema change request on the rw port is rejected.
File storage
A shared drive where agents and people keep files, with folders and sharing links that expire (by default after one week).
| Port | Tools |
|---|---|
| rw (Read & write files) | listFiles, getFile, putFile, deleteFile, moveFile, shareFile, unshareFile |
| ro (Read files) | listFiles, getFile |
Skill library
Operating instructions (procedures, checklists, style, with attached files) that the agent loads only when needed. It keeps the one-line list of skills in context and reads the relevant one in full: a long library costs nothing until it is used.
- The format is Agent Skills (a SKILL.md file with a name and description), the same used by Claude.
- Each skill's description decides when it opens: it must say when it applies.
- With tags, a library can have a shelf for each agent.
- A procedure placed in the instructions is reread on every turn and is not reusable; as a skill it is.
Related pages
How to create a new agent
Step by step: project, agent node, instructions, model, tools, channel, testing and publishing.
Code and integrations
TypeScript connectors, sandboxes, remote and hosted MCP servers: how to connect agents to business systems.
Security
Roles, groups, credentials, enterprise sign-in and least privilege: how to protect an agent system.
Have an agent to take to production?Let’s talk.
We show you TeraBrain NX on your case: systems to connect, access and controls.