Counting visitors…
Browse all docs
Start · 9Find your path through RunifoldLearn Runifold in 45 minutesUnderstand the complete Runifold platformYour first trustworthy runChoose the right execution APIChoose crates and Cargo featuresBuild common Runifold applicationsRunifold frequently asked questionsTroubleshoot Runifold applications
Execution kernel · 7Understand RunContextCoordinate external effects safelyBound work with budgets and cancellationHandle errors and retries safelyEvents, journals, and execution evidenceDesign capability-safe executionRecover safely from checkpoints
Models & providers · 7Route across models without duplicate outputChoose and configure a providerUse the provider-neutral model protocolBuild on the Provider Runtime contractUse OpenAI control-plane and Realtime APIsTest and benchmark provider adaptersSet up OpenAI, Anthropic, Gemini, and Ollama
Agents · 7Build and configure an AgentGive an Agent typed toolsAdd conversations and semantic memoryDelegate to child Agents safelyReturn structured Rust valuesStream without losing semanticsGround an Agent with retrieval
Durable workflows · 7Compose deterministic workflowsMake workflows durableOperate durable workflow workersCoordinate timers, signals, and durable waitsRun multi-tenant workflow infrastructureRun parallel branches and safe racesVersion and evolve durable workflows
Integrations · 7Connect through MCPChoose stores and persistence boundariesExpose durable work through MCP TasksBuild and evaluate retrieval pipelinesUse MCP Resources, Prompts, and SamplingCache MCP responses without crossing authorityDeploy Runifold in a Rust web service
Quality & operations · 10Test without the networkEvaluate quality and prevent regressionsObserve the complete run treeRun safely in browsers and at the edgeRead reliability claims preciselyRun reproducible evaluations in CIOperate Runifold with SLOsGovern Task retention and deletionArchive audit evidence to S3-compatible WORM storageManage compatibility and trusted releases
Docs/Documentation
NEW TO RUNIFOLD?Build the complete mental model in 45 minutes
Documentation

Find your path through Runifold

Choose a learning path, understand the execution model, and know which guide to read next.

Practical guide·10 min
First time here?

Run the 5-minute Quickstart first. Then use the Cookbook for your product task or take the 45-minute core course for the complete mental model.

Explore the whole platform

Runifold is not only an Agent library. Start from the product surface you actually need:

SurfaceCore guides
execution kernelRunContext, capability security, checkpoint recovery
models and providersmodel protocol, Provider Runtime, provider testing
AgentsAgent construction, tools, conversation
durable workflowsWorkers, parallelism, versioning, tenancy
integrationsMCP context and Sampling, MCP Tasks, retrieval pipelines, storage
quality and operationsevaluation CLI, Task governance, SLOs, trusted releases

Read the complete platform architecture before designing a system that crosses more than one surface.

Choose a learning path

Runifold lets a small model call grow into a bounded, observable, recoverable system without switching to a second orchestration abstraction. You do not need to learn the whole runtime before starting.

If you want to…Start hereThen read
call a model from RustFirst trustworthy runChoose an execution API
choose and configure a model serviceProvider setupProvider Runtime, Provider testing
build a tool-using AgentBuild an AgentTyped tools, structured output
build a chat or assistantConversations and memoryRetrieval, streaming
coordinate specialized AgentsDelegationBudgets and cancellation
run long-lived business workWorkflowsDurable workflows, effects
expose a Rust APIDeploy a web serviceRunContext, observability
ship to productionTestingEvaluation, observability, reliability

If this is your first visit, complete the first-run tutorial before opening the reference-oriented guides. It establishes the vocabulary used everywhere else.

The execution model

Five ideas connect the entire API:

  1. Run — every execution has identity, a parent, a lifetime, and shared accounting.
  2. Event — the runtime records semantic state transitions separately from model-visible conversation.
  3. Capability — registering a callable says what exists; granting a capability says what this run may use.
  4. Effect — external writes have explicit recovery and idempotency semantics.
  5. Checkpoint — durable progress can be resumed without pretending an interrupted operation definitely failed.

This separation is the reason an ergonomic prompt_text call and a durable multi-worker workflow can share the same execution kernel.

From prototype to production

Add constraints when the product needs them:

StageAddWhy
Exploreprompt_textshortest path to visible text
Integratetyped output and Toolsstable application contracts
Boundexplicit RunContextbudgets, authority, deadline, cancellation
Resilientrouting and effectssafe recovery around uncertain networks
Durablecheckpoints and workflowssurvive restarts and long waits
Operatejournals, OTel, tests, evaluationsexplain and improve behavior

Do not add every layer by default. A single classification endpoint may never need a workflow; a payment workflow should not rely on a prompt to define its transitions.

Where to look

When a guide and the API reference appear to disagree, pin the crate version, check the reference for that version, and report the mismatch with a minimal reproduction.

For quick answers, use the FAQ. For failures, follow the Troubleshooting guide before opening an issue.