@exellix/job-dispatcher

@exellix/job-dispatcher

Graph job dispatch orchestration for Exellix graph runtimes.

Validate → normalize → compile → execute → diagnostics in one place. Hosts send a JSON request; this package owns the dispatch pipeline and engine wiring.

Published on npm as @exellix/job-dispatcher (currently 1.0.4).

Install

npm install @exellix/job-dispatcher

Requires Node >=20.19.0 || >=22.12.0.

Dependencies: @exellix/graph-engine, @exellix/ai-tasks.

Exports

Import	Use
@exellix/job-dispatcher	Main re-exports
@exellix/job-dispatcher/graph	dispatchGraphJob, dispatchGraphJobWithEngine, createExellixGraphRuntimeFromEnv, compileGraphDispatchPlan, resolveInvocationPlan
@exellix/job-dispatcher/studio	Studio compile/execute adapters, dispatchGraphJobWithEngine
@exellix/job-dispatcher/jobs	createJobsGraphDispatcher, buildJobsGraphDispatchRequest (used by @exellix/jobs)
@exellix/job-dispatcher/node	dispatchGraphNode — single-node runs
@exellix/job-dispatcher/matrix	createMatrixGraphExecuteAdapter — matrix batch hosts

Quick start (orchestration only)

import { dispatchGraphJob } from '@exellix/job-dispatcher/graph';

const result = await dispatchGraphJob(request, {
  compileGraph: myCompileGraphAdapter,
  executeGraph: myExecuteGraphAdapter,
});

Studio graph run

import { dispatchGraphJobWithEngine } from '@exellix/job-dispatcher/studio';
import { createExellixGraphRuntimeFromEnv } from '@exellix/job-dispatcher/graph';

const { exellixRuntime } = await createExellixGraphRuntimeFromEnv({ graphLoader });

const result = await dispatchGraphJobWithEngine(
  {
    mode: 'graph',
    graph: authoringDocument,
    runtime: { input: { vulnerabilityType: 'xss' } },
    jobId: 'job-abc',
  },
  {
    graphRuntimeEnvelope: { job, jobMemory, runtimeObjects },
    agentId: 'my-agent',
    exellixRuntime,
  },
);

Jobs worker / factory

import { createJobsGraphDispatcher } from '@exellix/job-dispatcher/jobs';

const { dispatchGraphRun } = createJobsGraphDispatcher({
  graphLoader,
  eventEmitter: activix?.eventEmitter,
});

const result = await dispatchGraphRun({
  graphId: 'graph-1',
  jobId: 'job-1',
  runtimeInput: { recordId: 'rec-1' },
  context: { jobMemory: resolvedJobMemory },
});

@exellix/jobs re-exports the same API for backward compatibility.

Runtime jobMemory

The dispatcher treats runtime.context.jobMemory as host-resolved runtime memory and forwards it unchanged into graph-engine execution as runtime.jobMemory. It validates and diagnoses whether the graph reads jobMemory, but it does not resolve Memorix, Hippox, linked records, or associated snapshot data itself.

For jobs that assemble Memorix snapshot records, @exellix/jobs resolves those host concerns before dispatch (worker, executeGraphJob, Graph Studio BFF — never inside this package):

• runtime.input receives the graph's primary record fields.
• runtime.context.jobMemory.context["linked-information"] and extendedInformation receive configured linked/same-object context source enrichment (resolveJobMemoryForDispatch / worker context resolution).
• runtime.context.jobMemory.associated* fields receive associated/enriched sub-objects promoted out of snapshot data.

Graph Studio and other sync hosts should call @exellix/jobs with the same contextSources definition as queued jobs, then pass the resolved jobMemory into dispatchGraphJobWithEngine or createJobsGraphDispatcher().dispatchGraphRun.

Snapshot-associated data should be promoted into jobMemory, not left in the main runtime input and not nested under jobMemory.context. On snapshots, any root data property whose name starts with associated is treated as associated content, so this supports associatedData, associatedInferred, associatedAnalysis, associatedRiskScores, associatedExploitability, and other custom associated<Something> fields without hard-coding the known names. The worker normalizes each associated value to an array of object items and dispatches:

await dispatchGraphRun({
  graphId: 'graph-1',
  jobId: 'job-1',
  runtimeInput: {
    recordId: 'rec-1',
    cveId: 'CVE-2024-1234',
  },
  context: {
    jobMemory: {
      context: {
        'linked-information': [
          { context: 'Existing linked context', data: [{ recordId: 'linked-1' }] },
        ],
      },
      associatedData: [
        { owner: 'team-x', severity: 'critical' },
      ],
      associatedRiskScores: [
        { score: 91, source: 'risk-model-v2' },
      ],
    },
  },
});

Graphs or skill templates that need this snapshot-associated content should read it from direct jobMemory.associated* fields such as jobMemory.associatedData or jobMemory.associatedRiskScores. This keeps the business input clean while still making associated/enriched data available to skills that opt into jobMemory, and keeps jobMemory.context dedicated to context-source/Hippox-style context.

Single-node run

import { dispatchGraphNode } from '@exellix/job-dispatcher/node';

await dispatchGraphNode(
  { graph, nodeId, job, jobMemory, execution, variables },
  { exellixRuntime },
);

Compile-only preview

import { compileGraphDispatchPlan } from '@exellix/job-dispatcher/graph';

const compiled = await compileGraphDispatchPlan(request, {
  agentId,
  jobTypeId,
  graphRuntimeEnvelope,
});

What this package owns

• Request validation, normalization, orchestration, diagnostics
• Studio + jobs compile/execute adapters
• Runtime factory (createExellixGraphRuntimeFromEnv)
• Node and matrix execute facades
• Forwarding host-provided jobMemory into graph runtime and warning when assembled context is unused or missing

What stays in hosts

• Memorix / Hippox context resolution (@exellix/jobs)
• Snapshot record assembly, including promoting associated/enriched associated* data into direct jobMemory.associated* fields
• Queue admission, HTTP handlers
• Authoring UI (@exellix/graph-engine/authoring)

Development

npm run build --workspace=@exellix/job-dispatcher
node --test jobs-packages/job-dispatcher/tests/*.test.mjs
npm publish -w @exellix/job-dispatcher