> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.rightbrain.ai/v-1/docs/getting-started/introduction/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.rightbrain.ai/_mcp/server.

# Build reliable AI agents that run inside your existing tools

> Rightbrain is a platform for building reliable AI agents that run inside your existing tools and workflows, and operating them at scale.

Rightbrain gives you agents that run on the model of your choice, primitives you make available to them, triggers to run them inside your systems, and the production controls that make it all dependable at scale: approvals, evals, fallbacks, versioning, and a tamper-evident audit log. Build reliable agents, then operate them with every run attributed and traceable back to the exact prompt and model that produced it.

Use a Task when one model operation with typed output is enough. Use an agent when the work must decide, branch, use tools, or continue across turns. Both run through the same API-first platform and produce observable, revision-linked run records.

#### [Create your first Task](/docs/quickstart/create-a-task)

Build the smallest production unit, run it, and inspect its result.

#### [How Rightbrain works](/docs/getting-started/how-it-works)

Choose a Task or agent, then add only the primitives the work needs.

#### [Build your first agent](/docs/quickstart/first-agent)

Make a Task available to an agent and stream the run.

## The four primitives

An **agent** is the top-level unit: a goal-directed reasoner on the model of your choice that works out of the box with nothing attached. The power comes from the primitives you make available to it, each created, managed, and versioned independently of the agents that use it — at run time the agent picks the ones the job needs.

And when the job is a single, well-defined operation that must produce structured output reliably, skip the agent and use a **task** directly — a task is, in effect, a deterministic agent that does one thing very well, on its own model.

#### [Tasks](/docs/concepts/tasks)

Structured AI functions: a prompt with typed inputs and a typed output on a model you choose. Agents call them as tools, and you can run them standalone via API.

#### [Skills](/docs/concepts/skills)

Reusable, versioned abilities agents activate for focused work. A Skill can declare and provision its executable dependencies.

#### [Collections](/docs/concepts/collections)

Document and file knowledge bases for retrieval. A task retrieves from a collection through its RAG configuration.

#### [Connections](/docs/concepts/connections)

External reach: native integrations like HubSpot, Salesforce, Slack, Google Sheets, and Gmail, plus MCP servers.

## Production-ready by default

The hard part of AI is not the demo, it's everything around it. Rightbrain treats operations as a first-class concern.

#### [Approvals](/docs/production/approvals)

Require human sign-off before an agent runs a sensitive tool.

#### [Evals](/docs/production/evals)

Measure agent and task quality against a fixed set before you ship.

#### [Fallbacks](/docs/production/fallbacks)

Route to a fallback model when the primary one fails.

#### [Observability](/docs/production/observability)

Inspect every run: telemetry, failure reasons, and a tamper-evident audit log.

## Keep going

#### [How Rightbrain works](/docs/getting-started/how-it-works)

The mental model: agents, the primitives you make available to them, and how a run flows.

#### [Why Rightbrain](/docs/getting-started/why-rightbrain)

Why most AI proofs-of-concept never reach production, and how this changes that.

> Build reliable AI agents that run inside your existing tools and workflows. Rightbrain developer documentation.