Counting visitors…
RUNIFOLD
中文GitHub
Menu
QuickstartCookbookPlatformProvidersOperations
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/Help
NEW TO RUNIFOLD?Build the complete mental model in 45 minutes
Help

Runifold frequently asked questions

Get concise answers about APIs, providers, Runs, Tools, state, reliability, production use, and versions.

Practical guide·10 min

Starting and choosing

Is Runifold only an Agent framework?
No. Agent is one surface over the same execution kernel. You can use the model protocol, Provider Runtime, typed Tools, retrieval, MCP, evaluation, observability, or durable Workflow runtime independently.

Should I depend on runifold or a narrow crate?
Start with the runifold facade and enable only required features. Use narrow crates when you are building an adapter, minimizing a specialized dependency boundary, or publishing reusable infrastructure.

Which API should I start with?
Use prompt_text for one text result, prompt when you need the full outcome, stream for incremental events, and run with an explicit RunContext at production boundaries. See Choosing an API.

Providers and models

Can I switch providers without changing application code?
Core requests, Runs, Agents, and Tools are provider-neutral, but capabilities are not identical. Switch the client and model configuration, then rerun capability and quality verification.

Does OpenAI-compatible mean identical behavior?
No. It means an endpoint resembles a wire protocol. Streaming event shape, Tool behavior, structured output, usage, reasoning, and errors can differ.

Why does Runifold not read credentials automatically?
Applications own secret ingress. Explicit construction makes server, test, gateway, browser, and multi-tenant boundaries visible and prevents a library from silently reading ambient authority.

Runs, Tools, and state

What is the difference between an Agent and a Run?
An Agent describes model behavior, Tools, and turn policy. A Run is one execution tree with identity, budget, deadline, cancellation, capabilities, and journal state.

Does registering a Tool authorize it?
No. Registration says the Tool exists. The Run capability set decides whether this execution may call it.

Is conversation history the same as a checkpoint?
No. Conversation history is model context. A checkpoint records recoverable execution state. An event journal records what happened.

Reliability and production

Does Runifold guarantee exactly-once external effects?
No general runtime can guarantee that for an arbitrary remote system. Runifold coordinates write-ahead records, idempotency keys, result replay, and conservative recovery so applications can handle ambiguity explicitly.

When should I use a Workflow instead of an Agent?
Use an Agent when the model chooses the next model/Tool turn. Use a Workflow when the application owns ordered steps, branches, waits, recovery, or durable task state.

Is Runifold production-ready?
The project is pre-1.0. Production adoption should pin the release, verify the exact Provider/model/features in your environment, and use the published reliability evidence instead of assuming every surface has equal maturity.

Versions and support

Which version do these docs describe?
The current site and compile-checked examples target runifold 0.9.0.

How should I upgrade before 1.0?
Pin an exact minor release, read the changelog and migration notes, update all Runifold crates together, compile examples, run deterministic tests, then run a narrow live Provider smoke test.

For build or runtime failures, continue to Troubleshooting.