Workflows

The workflow runtime routes start runs of a deployed workflow, read them, and recover them when they park. Definitions are authored with lua push workflow and activated with lua workflows deploy; these routes only run them. This page covers the run model, the definition and run routes, recovery, and the error codes. Steps, events, artifacts, and exports are on Workflow events; approvals, signals, and gates on Workflow approvals; compose, schedules, goals, workspaces, and organization-wide routes on Workflow schedules and goals. Every route is the REST twin of a Workflows method or a lua workflows action, named under each endpoint. The CLI equivalents on this page require lua-cli 3.36.0 or later.

Base URL and authentication

Requests go to https://api.heylua.ai with Authorization: Bearer <<YOUR_API_KEY>>; the REST API overview covers keys, the typed error envelope, cursor pagination, and the per-minute rate buckets. Per-agent routes live under /workflows/:agentId and bind their scope to that agent: workflows:read for every GET, workflows:execute to start runs and act on them, workflows:write to compose, schedule, create or edit goals, and start a run from an inline graph or script. Reading step inputs and outputs needs workflows:read-outputs; without it payload fields come back as { "__hidden": true } and event data is redacted. Organization-wide routes bind to the orgId you pass. Approving, denying, editing an approval payload, resolving a parked step, erasing a run, and reassigning assets are human-only: a typed API credential is refused with 403 APPROVAL_REQUIRES_HUMAN, so drive those from a user session or the CLI. Where an endpoint names a Workflows method, the method works in deployed agents and under lua test alike, except Workflows.raiseBudget, which no runtime serves; the deployed members return less than the typings declare (see Deployed runtime differences).

Run model

A run is identified by runId (wfr_...), pins the workflowVersionId it started on, and finishes on that version: deploying a new version never moves a live run. It moves through these statuses:

Group	Statuses	Meaning
In flight	`queued`, `running`, `cancellation_requested`	The run holds an execution slot
Idle	`gated`, `suspended`, `waiting`	Nothing executes until a person, a signal, or a timer wakes it
Terminal	`completed`, `failed`, `cancelled`, `abandoned`, `timed_out`	Final; late writes are fenced off

abandoned is a run a person or an agent force-canceled once forceAvailableAt had passed; timed_out is a run that reached budget.maxDurationSeconds, idle or in flight. A gated or suspended run carries gate { kind, since, expiresAt?, stepId?, attempt?, code? } with kind one of start-consent (approvalLinkId), quota (code concurrency_cap or suspended_cap), billing (code, operationId), budget (dimension, spent, cap), exception, org_archived, disabled, or model_policy (code, stepId: a step whose model request the organization’s policy cannot resolve, parked rather than failed); Gates says which verb clears each. trigger records who started the run: chat, sdk, api (CLI and HTTP), schedule, webhook (a trigger primitive), template, workflow (a parent run), or device. origin records what it ran: inline, compose, definition, schedule, or goal. A run summary, the row every list returns, carries runId, workflowId, workflowName, workflowVersionId, form (graph or script), agentId, orgId, status, gate, trigger, origin, createdBy { subjectType, subjectId }, counts { total, completed, failed, skipped, running, suspended, pending, cancelled? }, createdAt, startedAt, updatedAt, completedAt, durationMs (terminal only), correlationKey, tags, principalKind (user, service, or customer), batchId, goalId, cancel, usage, error { code, message, stepId?, issues? }, hasOutput, restricted, stepCount, currentLabel, phase, and eventSeq. Timestamps are Unix time in milliseconds. The detail adds input, output (inline up to 256 KB, else { "__cdnRef" } beside an outputPreview and an outputRef), steps[], edges[], suspensions[],

budget { maxCredits?, maxSteps, maxDurationSeconds, deadlineAt, unit, spent { credits, actionsEstimate, steps }, reserved, remaining? }

, budgetRaises[], and, on a run whose steps were assigned models, modelConsent { plan, card }: the per-step model plan and the rendered lines of the card a person consents to. An agent step row carries taskClass, modelRequested, modelResolved, modelReason, effortApplied, and modelFallback[], each absent when the step does not have it; see Workflow events. The detail of a run that needed consent to start also carries consent { approvedBy, at, approvalLinkId?, via?, envelope? }. via is human when a person approved it and autonomy when the organization’s pre-consent envelope admitted it; an autonomous start has no approvalLinkId, records system:policy as approvedBy, and carries the resolved envelope that let it through.

Definitions

GET /workflows/:agentId/definitions

Lists the workflow definitions on an agent.

string

required

The agent.

string

graph or script.

boolean

default:"false"

Include chat-composed workflows.

string

nextCursor from the previous page.

integer

default:"50"

From 1 to 100.

Response 200 with { "items": [...], "nextCursor": string | null }; each item is

{ workflowId, name, description?, form, dynamic, active, activeVersionId?, inputSchema?, updatedAt, scheduleJob?, idleExpiresAt? }

. Equivalent: lua workflows list.

lua workflows list

const response = await fetch('https://api.heylua.ai/workflows/<<YOUR_AGENT_ID>>/definitions?limit=50', {
  headers: { Authorization: 'Bearer <<YOUR_API_KEY>>' },
});
const page: { items: Array<{ workflowId: string; name: string; active: boolean }>; nextCursor: string | null } =
  await response.json();
for (const definition of page.items) console.log(definition.name, definition.active);

curl "https://api.heylua.ai/workflows/<<YOUR_AGENT_ID>>/definitions?limit=50" \
  -H "Authorization: Bearer <<YOUR_API_KEY>>"

GET /workflows/:agentId/definitions/:workflowId/rollup

Returns a per-day rollup of one workflow’s runs, bucketed by the UTC day the run was created.

string

Start of the range, as Unix milliseconds or an ISO 8601 timestamp; defaults to 30 days before to.

string

End of the range, same formats; defaults to now.

Response 200 with { "workflowId", "bucket": "day", "from", "to", "rows": [...] }; each row is { date, runs, completed, failed, cancelled, suspended, p50DurationMs, p95DurationMs, creditsTotal, stepFailures }, where date is YYYY-MM-DD, failed counts failed, timed_out, and abandoned runs, suspended counts suspended, gated, and waiting runs, and the percentiles are null on a day with no terminal run. The answer is cached for 60 seconds. 400 ROLLUP_RANGE_TOO_WIDE (maxDays, days) past 90 days, or VALIDATION_FAILED for a timestamp that does not parse or a from after to; 404 WORKFLOW_NOT_FOUND.

POST /workflows/:agentId/definitions/:workflowId/recompose

Rewrites a chat-composed workflow into the model-class form: reads its active version, turns every agent step whose stored task class maps to the class its current pin heads into model: "class/<class>", and publishes the result as a new version on the same definition. No model call is made, so the rewrite is reproducible and the version diff reviewable; a pin that differs from the class map is a deliberate override and is reported under skipped. Needs workflows:write.

boolean

true or 1: answer the same diff and write nothing.

Response 200 with { "workflowId", "fromVersionId", "fromVersion", "dryRun", "rewritten": [...], "skipped": [...], "composed"? }: the version the rewrite was computed from, the steps it rewrote and skipped, and composed only when a version was published (never on a dry run, and not when nothing was rewritten). 404 WORKFLOW_NOT_FOUND; 409 RECOMPOSE_NOT_APPLICABLE with reason one of not-dynamic (a pushed workflow), no-active-version, or no-graph; 503 MODEL_POLICY_UNVERIFIABLE or CONTROL_UNAVAILABLE. Equivalent: lua workflows recompose <workflow>.

Run routes

POST /workflows/:agentId/runs

Starts one run.

string

required

The agent.

string

Replay key; wins over the body field. Only this spelling is read: the channel routes’ X-Idempotency-Key is ignored here, so a retry that sends it creates a second run.

string

The saved definition, by name (at most 120 characters). Send exactly one of workflowName, workflowId, composeId, graph, or script.

string

The saved definition, by id.

string

A draft from POST .../compose.

object

An inline dynamic graph, at most 256 KB; needs workflows:write.

string

An inline script, at most 256 KB; needs workflows:write.

any

The run input, at most 256 KB; validated against the definition’s inputSchema.

string

At most 128 characters. The same key with the same definition and input replays the existing run; anything else answers 409 IDEMPOTENCY_KEY_REUSED.

object

{ maxCredits?, maxSteps?, maxDurationSeconds?, maxJobSeconds? }, each an integer of at least 1, capped by the organization’s policy. Omit a key rather than sending 0.

integer

From 0 to 55. Long-polls: when the run settles within the window the answer is 200 with the run detail instead of 202.

string

Pin a version; defaults to the active one.

string

A caller-chosen, non-unique key (at most 128 characters of letters, digits, :, _, -, ., /) for signals/by-key and list filters.

string[]

At most 10 tags of at most 40 characters.

string

default:"app"

Where the terminal summary goes: emailApp, email, app, or off.

object

{ channel, threadId } with channel one of whatsapp, sms, email, webchat, slack: the terminal summary is delivered on that channel. On slack it is posted through the agent’s own Slack bot; threadId is any Slack conversationId that Channels.send takes, usually C…:threadTs, or a Lua.request.conversation.externalId when the agent has more than one Slack connection. The run’s occurrence id is the idempotency key, so a redelivery never posts twice. With no Slack connection, or the bot not in the channel, the reply fails permanently (channel_unsupported, or thread_not_found for a thread Slack doesn’t know). Attachments go out as links.

string

For chat-started runs, the caller’s own thread on this agent where the run card lands (at most 256 characters); any other thread answers 400 ORIGIN_THREAD_INVALID.

boolean

Take one background agent turn on originThreadId when the run ends; without an origin thread the start is refused with 400 CALLBACK_THREAD_REQUIRED.

object

{ kind: "customer", externalId }: the run belongs to a customer, who can never be an approver.

object

{ repo?, ref?, credentialsRef?, sizeGb? } overriding a declared Job-tier workspace; sizeGb from 1 to 50.

object

Seeds the run state; at most 64 KB.

Response 202 with { "runId", "status": "queued" | "gated", "watchHint": { "events", "poll" } }. An idempotent replay answers 200 with the existing run’s runId, status, workflowId, workflowName, startedAt (ISO 8601), watchHint, and idempotentReplay: true, deduplicated: true; nothing was created. When waitSeconds elapsed on a settled run the answer is 200 with the run detail. Errors

Status	Code	Meaning
`400`	`VALIDATION_FAILED`, `ORIGIN_THREAD_INVALID`, `CALLBACK_THREAD_REQUIRED`, `CORRELATION_KEY_INVALID`	Bad body; an inline definition answers the compose validator’s issues, each with a `path`
`402`	`INSUFFICIENT_CREDITS`	The organization cannot pay for the run
`403`	`INSUFFICIENT_SCOPE`	An inline `graph` or `script` without `workflows:write`
`404`	`WORKFLOW_NOT_FOUND`	No such definition on this agent
`409`	`IDEMPOTENCY_KEY_REUSED`, `RUNS_IN_FLIGHT`	The key is held by another start (`existingRunId`); the workflow’s `concurrencyPolicy` is `forbid` and a run is live (`blockingRunId`, `blockingStatus`)
`413`	`PAYLOAD_TOO_LARGE`	`input` or `graph` over 256 KB
`422`	`COMPOSE_FORM_FORBIDDEN`	An inline `script` under a graph-only policy
`429`	`CONCURRENCY_CAP`, `SUSPENDED_CAP`	The organization’s cap is full (`cap`, `active` or `suspended`, `retryAfterSeconds`)
`501`	`INLINE_START_UNSUPPORTED`	The server cannot validate the inline definition; nothing started, do not retry unchanged
`503`	`CONTROL_UNAVAILABLE`	The control plane did not answer; `ambiguous: true`, retry with the same key

Equivalent: lua workflows start <workflow> --input @input.json --follow, Workflows.start().

lua workflows start lead-outreach --input '{"segment":"trial"}' \
  --idempotency-key outreach-2026-09-12 --correlation-key outreach:2026-09-12 --tag nightly

const response = await fetch('https://api.heylua.ai/workflows/<<YOUR_AGENT_ID>>/runs', {
  method: 'POST',
  headers: {
    Authorization: 'Bearer <<YOUR_API_KEY>>',
    'Content-Type': 'application/json',
    'Idempotency-Key': 'outreach-2026-09-12',
  },
  body: JSON.stringify({
    workflowName: 'lead-outreach',
    input: { segment: 'trial' },
    correlationKey: 'outreach:2026-09-12',
    tags: ['nightly'],
  }),
});
const started: { runId: string; status: string; idempotentReplay?: boolean } = await response.json();
console.log(response.status, started.runId, started.status);

curl -X POST "https://api.heylua.ai/workflows/<<YOUR_AGENT_ID>>/runs" \
  -H "Authorization: Bearer <<YOUR_API_KEY>>" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: outreach-2026-09-12" \
  -d '{ "workflowName": "lead-outreach", "input": { "segment": "trial" }, "correlationKey": "outreach:2026-09-12", "tags": ["nightly"] }'

POST /workflows/:agentId/runs/batch

Starts many runs of one workflow at once. The body may be at most 2 MB.

string

required

The definition.

array

required

From 1 to 200 items, subject to the organization’s batch cap: { idempotencyKey (required, at most 128 characters, unique within the batch), input?, budget?, tags?, correlationKey? }.

string

default:"gate"

gate parks items over the concurrency cap; reject refuses them.

object

Batch-wide default; an item’s own budget wins.

string

Pin a version.

Response 202 with { "batchId", "accepted", "items": [{ idempotencyKey, runId?, status, code?, gate?, issues? }] }, items in the order sent; status is queued, gated, replayed, or rejected, and accepted counts the first two. Errors 400 BATCH_TOO_LARGE (max, count) or BATCH_ITEM_INVALID (index, issues; also a duplicate idempotencyKey); 404 WORKFLOW_NOT_FOUND; 409 RUNS_IN_FLIGHT under forbid, or ORG_PURGING, either of which refuses the whole batch; 413 PAYLOAD_TOO_LARGE; 503 CONTROL_UNAVAILABLE, retry with the same keys. Equivalent: Workflows.startBatch().

GET /workflows/:agentId/runs

Lists runs on an agent, newest first by default. Rows carry metadata and counts, never outputs.

string

Comma-separated statuses.

string

Only runs of this definition.

string

One trigger value.

integer

Unix milliseconds; runs updated at or after it.

string

Runs of one batch start.

string

Exact match.

string

Repeatable, at most 10.

integer

Terminal runs only; maxDurationMs likewise.

string

default:"-createdAt"

-createdAt, createdAt, -durationMs, or durationMs.

string

nextCursor from the previous page.

integer

default:"50"

From 1 to 100.

Response 200 with { "items": [...run summaries], "nextCursor": string | null }. 400 BAD_CURSOR, CORRELATION_KEY_INVALID, or VALIDATION_FAILED with the offending parameter in issues[].path. Equivalent: lua workflows runs --workflow <name> --status failed, Workflows.list().

lua workflows runs --status failed --tag nightly --limit 20

const params = new URLSearchParams({ status: 'failed,timed_out', tag: 'nightly', limit: '20' });
const response = await fetch(`https://api.heylua.ai/workflows/<<YOUR_AGENT_ID>>/runs?${params}`, {
  headers: { Authorization: 'Bearer <<YOUR_API_KEY>>' },
});
const page: { items: Array<{ runId: string; status: string; workflowName?: string }>; nextCursor: string | null } =
  await response.json();
for (const run of page.items) console.log(run.runId, run.status);

curl -G "https://api.heylua.ai/workflows/<<YOUR_AGENT_ID>>/runs" \
  -H "Authorization: Bearer <<YOUR_API_KEY>>" \
  --data-urlencode 'status=failed,timed_out' --data-urlencode 'tag=nightly' --data-urlencode 'limit=20'

GET /workflows/:agentId/runs/:runId

Returns one run.

string

default:"summary"

full adds inputs and outputs for callers who may read them, plus receipt and notify for the creator or an org:manage holder.

Response 200 with the run detail. 404 RUN_NOT_FOUND; any other value of fields is 400 VALIDATION_FAILED. Equivalent: lua workflows status <runId> --steps, Workflows.get().

lua workflows status <runId> --steps

const response = await fetch('https://api.heylua.ai/workflows/<<YOUR_AGENT_ID>>/runs/<<RUN_ID>>?fields=full', {
  headers: { Authorization: 'Bearer <<YOUR_API_KEY>>' },
});
const run: { status: string; output?: unknown; steps: Array<{ stepId: string; status: string }> } = await response.json();
console.log(run.status, run.steps.map((step) => `${step.stepId}=${step.status}`).join(' '));

curl "https://api.heylua.ai/workflows/<<YOUR_AGENT_ID>>/runs/<<RUN_ID>>?fields=full" \
  -H "Authorization: Bearer <<YOUR_API_KEY>>"

DELETE /workflows/:agentId/runs/:runId

Erases a terminal run with its events, steps, signals, and artifacts. Creator or organization admin (org:manage), person only. Response 202 with { "accepted": true, "purgeId" }. 403 APPROVAL_REQUIRES_HUMAN or NOT_RUN_CREATOR; 404 RUN_NOT_FOUND; 409 RUN_NOT_TERMINAL with nextAction: "cancel"; 501 ERASE_UNAVAILABLE; 503 CONTROL_UNAVAILABLE. Equivalent: lua workflows delete-run <runId> --yes.

Recovery

POST /workflows/:agentId/runs/:runId/cancel

Asks a run to stop at the next step boundary, and escalates to a force cancel when asked again after forceAvailableAt. Creator or a workflows:execute holder.

string

default:"request"

request or force.

string

At most 2,000 characters, kept for audit.

Response 200 with the verdict

{ "cancelRequested", "state", "status", "transitioned", "requestedAt"?, "nextAction": "cancel_again" | "force" | "none", "forceAvailableAt"?, "forced"? }

. A force before forceAvailableAt is a 200 with transitioned: false and nextAction: "cancel_again", and a terminal run is a 200 with state: "terminal" and nextAction: "none"; neither is an error. 403 NOT_RUN_CREATOR; 404 RUN_NOT_FOUND; 503 CONTROL_UNAVAILABLE. Equivalent: lua workflows cancel <runId> --reason "wrong input", Workflows.cancel().

lua workflows cancel <runId> --reason "wrong input"

const response = await fetch('https://api.heylua.ai/workflows/<<YOUR_AGENT_ID>>/runs/<<RUN_ID>>/cancel', {
  method: 'POST',
  headers: { Authorization: 'Bearer <<YOUR_API_KEY>>', 'Content-Type': 'application/json' },
  body: JSON.stringify({ reason: 'wrong input' }),
});
const verdict: { status: string; nextAction: 'cancel_again' | 'force' | 'none'; forceAvailableAt?: number } =
  await response.json();
console.log(verdict.status, verdict.nextAction);

curl -X POST "https://api.heylua.ai/workflows/<<YOUR_AGENT_ID>>/runs/<<RUN_ID>>/cancel" \
  -H "Authorization: Bearer <<YOUR_API_KEY>>" \
  -H "Content-Type: application/json" \
  -d '{ "reason": "wrong input" }'

POST /workflows/:agentId/runs/:runId/steps/:stepId/retry

Re-arms a step that failed past its retries, parked on an exception gate, or is on a billing hold.

string

At most 2,000 characters.

boolean

Restart the step’s automatic retry budget; the attempt number keeps counting up. Ignored on a gate or billing park.

Response 200 with { "retried": true, "attempt", "runStatus", "remainingParks" }, or the no-op { "retried": false, "reason": "already_retried" | "already_resolved", "recorded": { at?, by?, outcome? }, "runStatus" } when a decision already closed the park. 404 RUN_NOT_FOUND or STEP_NOT_FOUND; 409 STEP_NOT_PARKED (stepId, status), STEP_RETRY_CAP (a retry that would run past attempt 20: stepId, attempt, toAttempt, maxAttempts), or RUN_TERMINAL. Equivalent: lua workflows retry-step <runId> --step <stepId> --note "vendor back up".

POST /workflows/:agentId/runs/:runId/steps/:stepId/resolve

Resolves a parked step by skipping it, completing it with an output you supply, or failing it. Person only.

string

default:"complete"

skip, complete, or fail.

any

Required for complete, ignored otherwise; at most 256 KB serialized; validated against the step’s outputSchema.

string

At most 2,000 characters.

Response 200 with { "resolved": true, "outcome", "runStatus", "remainingParks" }, or the no-op { "resolved": false, "reason": "already_resolved" | "already_retried", "recorded", "runStatus" }. 400 VALIDATION_FAILED (output-required) or RESOLVE_OUTPUT_INVALID (stepId, issues); 403 APPROVAL_REQUIRES_HUMAN or NOT_RUN_CREATOR; 404 RUN_NOT_FOUND or STEP_NOT_FOUND; 409 STEP_NOT_PARKED or RUN_TERMINAL; 413 OUTPUT_TOO_LARGE. Equivalent: lua workflows resolve-step <runId> --step <stepId> --outcome complete --output '{"sent":0}'.

lua workflows resolve-step <runId> --step sendEmails --outcome skip --note "Emails were sent by hand"

const response = await fetch('https://api.heylua.ai/workflows/<<YOUR_AGENT_ID>>/runs/<<RUN_ID>>/steps/sendEmails/resolve', {
  method: 'POST',
  headers: { Authorization: 'Bearer <<YOUR_API_KEY>>', 'Content-Type': 'application/json' },
  body: JSON.stringify({ outcome: 'skip', note: 'Emails were sent by hand' }),
});
const resolved: { resolved: boolean; outcome?: string; runStatus: string } = await response.json();
console.log(resolved.runStatus);

curl -X POST "https://api.heylua.ai/workflows/<<YOUR_AGENT_ID>>/runs/<<RUN_ID>>/steps/sendEmails/resolve" \
  -H "Authorization: Bearer <<YOUR_API_KEY>>" \
  -H "Content-Type: application/json" \
  -d '{ "outcome": "skip", "note": "Emails were sent by hand" }'

POST /workflows/:agentId/runs/:runId/budget

Raises the budget of a run parked on a budget gate. Creator or organization admin.

number

New credit cap, above 0.

integer

New step cap, at least 1.

integer

New Job-tier seconds cap, at least 1.

integer

New wall-clock cap, at least 1.

string

At most 2,000 characters.

Response 200 with { "raised": true, "budget", "runStatus", "resumed" }, or { "raised": false, "reason": "already_raised", "runStatus" }; resumed: false means the raise still does not cover the next step’s reserve and the run parked again. 400 VALIDATION_FAILED (no cap sent, or a cap at or below the current one) or CAP_EXCEEDED (above the organization’s ceiling); 403 NOT_RUN_CREATOR; 409 BUDGET_NOT_RAISABLE when the run is not parked on a budget. Equivalent: lua workflows raise-budget <runId> --credits 20. Workflows.raiseBudget() is typed but never works: deployed code gets not_implemented (501) and lua test gets WORKFLOWS_API_UNAVAILABLE.

Error codes

Every code the four workflow pages can answer, by status.

Status	Codes
`400`	`VALIDATION_FAILED`, `BAD_CURSOR`, `CORRELATION_KEY_INVALID`, `ORIGIN_THREAD_INVALID`, `CALLBACK_THREAD_REQUIRED`, `BATCH_TOO_LARGE`, `BATCH_ITEM_INVALID`, `RESUME_SCHEMA_INVALID`, `SIGNAL_SCHEMA_INVALID`, `PATCH_INVALID`, `RESOLVE_OUTPUT_INVALID`, `CAP_EXCEEDED`, `ARTEFACT_SCHEMA_INVALID`, `WORKFLOW_NOT_ON_AGENT`, `GOAL_MAX_RUNS_INVALID`, `GOAL_RAISE_BELOW_SPENT`, `REASSIGN_SAME_USER`, `ROLLUP_RANGE_TOO_WIDE`, `MODEL_NOT_ELIGIBLE`, `MODEL_CLASS_EMPTY`, `MODEL_CLASS_UNREACHABLE`, `MODEL_JOB_PROVIDER_UNSUPPORTED`
`402`	`INSUFFICIENT_CREDITS`
`403`	`INSUFFICIENT_SCOPE`, `NOT_RUN_CREATOR`, `APPROVAL_REQUIRES_HUMAN`, `READ_OUTPUTS_REQUIRED`, `NOT_AN_APPROVER`, `FOUR_EYES_REQUIRED`, `APPROVAL_SURFACE_DENIED`, `STEP_UP_REQUIRED`, `NOT_ITEM_APPROVER`, `CUSTOMER_PRINCIPAL_CANNOT_APPROVE`, `NOT_ORG_ADMIN`
`404`	`RUN_NOT_FOUND`, `STEP_NOT_FOUND`, `WORKFLOW_NOT_FOUND`, `AGENT_NOT_FOUND`, `WORKFLOWS_DISABLED`, `NOT_SCRIPT_RUN`, `APPROVAL_NOT_FOUND`, `ITEM_NOT_FOUND`, `CORRELATION_KEY_NOT_FOUND`, `SCHEDULE_NOT_FOUND`, `GOAL_NOT_FOUND`, `ARTEFACT_NOT_FOUND`, `EXPORT_NOT_FOUND`, `WORKSPACE_NOT_FOUND`, `JOB_NOT_FOUND`, `JOB_LOGS_NOT_PERSISTED`, `USER_NOT_IN_ORG`, `ORG_NOT_FOUND`
`409`	`IDEMPOTENCY_KEY_REUSED`, `RUNS_IN_FLIGHT`, `RUN_TERMINAL`, `RUN_NOT_TERMINAL`, `NOT_SUSPENDED`, `USE_SIGNAL_ROUTE`, `NOT_RESUMABLE`, `PAYLOAD_MISMATCH`, `EDIT_NOT_ALLOWED`, `PAYLOAD_FINGERPRINT_STALE`, `ITEM_ALREADY_DECIDED`, `CORRELATION_KEY_AMBIGUOUS`, `ORG_PURGING`, `SCHEDULE_CAP`, `GOAL_SCHEDULE`, `GOAL_CAP`, `GOAL_NOT_ACTIVE`, `GOAL_VERSION_CONFLICT`, `GOAL_BUDGET_EXHAUSTED`, `STEP_NOT_PARKED`, `STEP_RETRY_CAP`, `BUDGET_NOT_RAISABLE`, `ARTEFACT_PENDING`, `EXPORT_IN_PROGRESS`, `WORKSPACE_IN_USE`, `NOT_A_WORKFLOW_RUN`, `REASSIGN_TARGET_LACKS_EXECUTE`, `NOT_GATED`, `RECOMPOSE_NOT_APPLICABLE`
`413`	`PAYLOAD_TOO_LARGE`, `OUTPUT_TOO_LARGE`, `ARTEFACT_TOO_LARGE`, `PAYLOAD_PAGE_REQUIRED`
`422`	`COMPOSE_FORM_FORBIDDEN`, `SIGNAL_SCHEMA_UNCOMPILABLE`, `RESUME_SCHEMA_UNCOMPILABLE`
`423`	`ORG_ARCHIVED`
`429`	`RATE_LIMITED`, `CONCURRENCY_CAP`, `SUSPENDED_CAP`, `SSE_SUBSCRIBER_CAP`, `SIGNAL_INBOX_FULL`, `EXPORT_RATE_LIMITED`
`501`	`INLINE_START_UNSUPPORTED`, `ERASE_UNAVAILABLE`
`503`	`CONTROL_UNAVAILABLE`, `UPSTREAM_UNAVAILABLE`, `SCHEDULE_PROVISION_FAILED`, `ARTEFACT_READ_FAILED`, `MODEL_POLICY_UNVERIFIABLE`

SDK

CLI

REST API

Base URL and authentication

Run model

Definitions

GET /workflows/:agentId/definitions

GET /workflows/:agentId/definitions/:workflowId/rollup

POST /workflows/:agentId/definitions/:workflowId/recompose

Run routes

POST /workflows/:agentId/runs

POST /workflows/:agentId/runs/batch

GET /workflows/:agentId/runs

GET /workflows/:agentId/runs/:runId

DELETE /workflows/:agentId/runs/:runId

Recovery

POST /workflows/:agentId/runs/:runId/cancel

POST /workflows/:agentId/runs/:runId/steps/:stepId/retry

POST /workflows/:agentId/runs/:runId/steps/:stepId/resolve

POST /workflows/:agentId/runs/:runId/budget

Error codes

See also

​Base URL and authentication

​Run model

​Definitions

​GET /workflows/:agentId/definitions

​GET /workflows/:agentId/definitions/:workflowId/rollup

​POST /workflows/:agentId/definitions/:workflowId/recompose

​Run routes

​POST /workflows/:agentId/runs

​POST /workflows/:agentId/runs/batch

​GET /workflows/:agentId/runs

​GET /workflows/:agentId/runs/:runId

​DELETE /workflows/:agentId/runs/:runId

​Recovery

​POST /workflows/:agentId/runs/:runId/cancel

​POST /workflows/:agentId/runs/:runId/steps/:stepId/retry

​POST /workflows/:agentId/runs/:runId/steps/:stepId/resolve

​POST /workflows/:agentId/runs/:runId/budget

​Error codes

​See also

Base URL and authentication

Run model

Definitions

GET /workflows/:agentId/definitions

GET /workflows/:agentId/definitions/:workflowId/rollup

POST /workflows/:agentId/definitions/:workflowId/recompose

Run routes

POST /workflows/:agentId/runs

POST /workflows/:agentId/runs/batch

GET /workflows/:agentId/runs

GET /workflows/:agentId/runs/:runId

DELETE /workflows/:agentId/runs/:runId

Recovery

POST /workflows/:agentId/runs/:runId/cancel

POST /workflows/:agentId/runs/:runId/steps/:stepId/retry

POST /workflows/:agentId/runs/:runId/steps/:stepId/resolve

POST /workflows/:agentId/runs/:runId/budget

Error codes

See also