# Brane

> Runtime control for AI agents. Write Python policies that govern tool calls, model calls, memory, retrieval, MCP tools, and other agent actions.

## Docs

- [AgentAction](https://docs.brane.membranelabs.org/concepts/agent-action.md): An AgentAction is one attempted use of a capability.
- [Capability](https://docs.brane.membranelabs.org/concepts/capability.md): A capability is anything an agent can use.
- [Decision](https://docs.brane.membranelabs.org/concepts/decision.md): A Decision is the output of policy evaluation.
- [Mental Model](https://docs.brane.membranelabs.org/concepts/mental-model.md): Capability + AgentAction + PolicyContext -> Decision. The Brane control loop.
- [PolicyContext](https://docs.brane.membranelabs.org/concepts/policy-context.md): The clean interface policy authors use to inspect an AgentAction.
- [Current Status](https://docs.brane.membranelabs.org/current-status.md): What is implemented in brane-core today and what is planned.
- [AI Agent Guardrails](https://docs.brane.membranelabs.org/guides/agent-guardrails.md): Use Brane as runtime guardrails for AI agents by enforcing policy before tool calls, model calls, database queries, MCP tools, and other agent actions execute.
- [Brane vs Prompt Guardrails](https://docs.brane.membranelabs.org/guides/brane-vs-guardrails.md): Brane differs from prompt guardrails by enforcing policy at the AI agent action boundary before tools, model calls, database queries, and MCP tools execute.
- [MCP Tool Governance](https://docs.brane.membranelabs.org/guides/mcp-tool-governance.md): Use Brane to govern MCP server and MCP tool access by treating MCP tools as capabilities with runtime policies.
- [OpenAI Agents SDK Guardrails](https://docs.brane.membranelabs.org/guides/openai-agents-sdk-guardrails.md): Use Brane as a runtime policy layer for OpenAI Agents SDK function tools and agent actions.
- [Policy-as-Code For AI Agents](https://docs.brane.membranelabs.org/guides/policy-as-code-for-ai-agents.md): Brane lets developers write Python policy functions that govern AI agent actions before capabilities execute.
- [AI Agent Tool Call Governance](https://docs.brane.membranelabs.org/guides/tool-call-governance.md): Govern AI agent tool calls with Brane by registering tools as capabilities and evaluating policies before each tool executes.
- [Brane Documentation](https://docs.brane.membranelabs.org/index.md): Govern every action your AI agents attempt.
- [After Capability Policies](https://docs.brane.membranelabs.org/policy/after-capability.md): Policies that run after a capability executes.
- [Before Capability Policies](https://docs.brane.membranelabs.org/policy/before-capability.md): Policies that run before a capability executes.
- [Writing Policies](https://docs.brane.membranelabs.org/policy/writing-policies.md): How to write before_capability and after_capability policies in brane-core.
- [Quickstart](https://docs.brane.membranelabs.org/quickstart.md): Install brane-core and govern your first tool capability.
- [Refund Limit Policy](https://docs.brane.membranelabs.org/recipes/refund-limit.md): Enforce a per-tenant refund limit with a before_capability policy.
- [SQL Read-Only Policy](https://docs.brane.membranelabs.org/recipes/sql-read-only.md): Block any SQL capability call that is not a SELECT statement.
- [Roadmap](https://docs.brane.membranelabs.org/roadmap.md): What is built, what is coming next, and the full Brane platform direction.
- [Runtime](https://docs.brane.membranelabs.org/sdk/runtime.md): The Runtime is the coordinator for capabilities and policies in brane-core.

## OpenAPI Specs

- [openapi](https://docs.brane.membranelabs.org/api-reference/openapi.json)