Mayura documentation

Mayura is a TypeScript framework for building AI agents, typed tools and durable workflows. New here? Read the introduction, then follow the quickstart.

#Get started

• Introduction: what Mayura is and how its parts fit together.
• Quickstart: create a project, or add a first agent to your own.
• Installation: requirements, optional packages and TypeScript settings.
• Using Mayura with AI coding agents: documentation your assistant can read offline.

#Concepts

• Agents: instructions, schemas, a model and tools.
• Tools: typed functions with declared effects.
• Runtime: running agents, events, cancellation and limits.
• Outcomes: how a run ends, error codes, and what to do about each.
• Permissions: the allow-list every run is checked against.
• Costs and budgets: prices, cost caps and shared budgets.
• Workflows: multi-step work, in memory or durable.

#Agents and models

• Model providers: OpenAI, Anthropic and OpenAI-compatible providers.
• Model routing: failover between providers.
• Streaming: show an answer while it is written.
• Vision: agents that see images and PDFs, from the input or from tools.
• Child agents: agents that delegate to other agents.
• Skills: SKILL.md folders an agent loads on demand.
• MCP tools: call tools on an MCP server.
• Terminal chat and commands: run an agent in a terminal.
• Testing: test agents, tools and workflows offline.

#Workflows

• Workflow composition: in-memory workflow graphs and workflows as tools.
• Durable workflows: steps that survive restarts.
• Approvals and human input: wait for people.
• Sagas and loops: compensation and repetition.
• Webhooks: start and signal workflows from signed webhooks.
• Operating workflows: pause, resume, cancel, versions and migrations.

#Data, safety and extensions

• Storage: SQLite and PostgreSQL.
• Memory and context: what an agent remembers and sees.
• Guardrails: check, block or redact inputs and outputs.
• Lifecycle hooks: code that can stop a run at defined points.
• Artifacts: store files that runs produce.
• Code Mode: run model-written code in a sandbox.
• Helpers: configuration, secrets, retries and other utilities.

#Serve, observe and deploy

• Server and client: serve agents and workflows over HTTP.
• React: hooks, components and forms.
• Observability: events, tracing and OpenTelemetry.
• Operator console: a UI for runs, workflows and approvals.
• Deployment: run Mayura in production.

#CLI

• Overview: the mayura command.
• mayura init: starters and templates.
• mayura dev: develop with rebuild and restart.
• serve, worker and migrate: run an application.
• Operator commands: manage a running server.

#Reference

• Entry points: every mayura/... import.

#Project

• Versioning and stability: what each release promises.
• Supported platforms: Node.js versions, operating systems and getting help.
• Security model: what Mayura protects against, and reporting vulnerabilities.
• Credits: the open-source projects and standards Mayura is built on.