> ## Documentation Index
> Fetch the complete documentation index at: https://docs.obversa.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# @obversa/runtime

> Compile graph definitions and record node outcomes.

`@obversa/runtime` compiles graph definitions and records graph decisions and
node outcomes. A host can resume unfinished positions from that record.

## Install

```bash theme={null}
pnpm add @obversa/runtime
```

Node.js 22.12 or later is required.

## Public entry points

From `@obversa/runtime` (`packages/runtime/src/api.ts`):

* **Graph definitions:** `compileGraph`, `dagGraphType`, and `convergence`
  define dependency graphs and repeated reviews.
* **Graph runs:** `createGraphExecutor`, `resolveGraphPlan`, and
  `persistRunDefinition` run and resume a stored graph.
* **In-process callbacks:** `createCallbackGate`, `createCallbackClient`,
  `replayCallbackClient`, and `directRouter` manage replayable questions.
* **Stored callbacks:** `createStoredCallbackClient` stores callback history in
  the run event stream.
* **Proof artifacts:** `writeProofArtifact` stores one stable JSON proof packet
  and returns its content digest.
* **Accepted results:** `createAcceptedResultRecord`, `resolveAcceptedResult`,
  and `acceptedResultMatches` bind a result to its reviewed inputs, proof,
  graph, workspace, and caller-supplied reviewer identity.
* **Approval:** `createApprovalCallbackGate`, `approvalSubjectDigest`, and
  `resolveApproval` bind a stored decision to the exact subject bytes. The
  stored client writes the callback submission and approval together.
* **Job helpers:** `run`, `loop`, `pipeline`, `dag`, and `sequence` are the
  older job API.

Subpaths:

* **Testing:** `@obversa/runtime/testing` provides graph definition helpers and
  mocks.
* **Local storage:** `@obversa/runtime/storage/local` provides
  `createLocalRunStorage`.

## Runnable example

From an Obversa checkout, run `pnpm example:pipeline`. The source is
`examples/packages/pipeline.ts`. It stores a three-stage pipeline and runs it
through `createGraphExecutor`.

See [Built-in pipeline](/graphs/pipeline) for the report that command prints.