Product / Work with agents

Assistant

Give a coding agent the same contract as your team, bounded by the same permissions.

Supported console operations are available to authorized agents through permission-scoped tools.Verified 2026-08-15 · owner GSB Core · evidence ↗

One tool registry

The CLI and the MCP server share one registry and implementation, so an agent cannot reach an operation the platform does not expose to you.

Bounded reads

Inspect exactly what an agent can see with a read-only tool list before connecting it to anything.

Approved writes

Writes, destructive operations, and external side effects require explicit approval rather than running on the model's word.

Scoped to the caller

An agent session acts inside the workspace permissions granted to it, enforced by the backend rather than by the prompt.

Operations

Each entry links to its reference page, generated from the same documentation the CLI and the MCP server return.

Keep reading

The guide covers the working shape of this capability, including request payloads, limits, and failure handling.

Rest of the platform

Database ↗

Query and transactionally save related data through one permission-aware entity contract.

Auth ↗

Enforce identity, tenancy, and record-level authorization at the data layer instead of in the interface.

Functions ↗

Add tenant-resident logic and shared libraries where data operations are not enough.

Workflows ↗

Describe durable, inspectable processes instead of hiding state in glue code.

Documents ↗

Turn governed multilingual content into stored files, PDFs, and email sent under your own identity.