AgentMeshStart free

API

Workflows API & execution modes

A workflow connects agents so one step’s output becomes the next step’s input. That single rule explains every mode below.

Execution modes

sequential    a chain — each step gets the previous step's output
parallel      every step gets the SAME input, run concurrently
conditional   branch on a field (equals, not_equals, greater_than,
              less_than, contains, exists, not_exists)
fan_out       run the SAME step once per item in a list, concurrently
loop          feed a step's output back into itself until a limit/condition
mixed         a dependency graph (DAG) via depends_on

Sequential workflows are available on the Free plan. Parallel, conditional, fan-out, loop and DAG require Pro — including for scheduled runs, which are skipped if the owner is not entitled.

Endpoints

POST   /v1/workflows                                    create
GET    /v1/workflows                                    list
GET    /v1/workflows/{id}                               fetch one
PUT    /v1/workflows/{id}                               update (new version)
DELETE /v1/workflows/{id}                               delete
POST   /v1/workflows/{id}/activate                      status -> ACTIVE
POST   /v1/workflows/{id}/pause                         pause
POST   /v1/workflows/{id}/execute                       run it
POST   /v1/workflows/{id}/validate                      validate the graph
GET    /v1/workflows/{id}/executions                    run history
GET    /v1/workflows/{id}/executions/{execId}           one run (+ steps)
POST   /v1/workflows/{id}/executions/{execId}/replay    re-run with same input
POST   /v1/workflows/{id}/executions/{execId}/cancel    cancel
GET    /v1/workflows/{id}/export                        export as JSON
POST   /v1/workflows/import                             import
GET    /v1/workflows/{id}/costs                         token + cost breakdown

Running a workflow

The body wraps your JSON in inputData. Whatever you pass becomes the workflow’s input, and flows into the first step.

curl -X POST https://app.agent-mesh.org/v1/workflows/$WORKFLOW_ID/execute \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"inputData": {"message": "The app crashed twice when exporting a report."}}'

Fan-out expects a list. Give it an items array and the same step runs once per item, concurrently:

{"inputData": {"items": ["Love the new dashboard.", "Support was slow.", "Works as described."]}}

Structured hand-off

Plain text output passes to the next step as message. If the upstream agent has Structured JSON output enabled, its fields become top-level input for the next step — which is what makes conditional branching reliable, because the branch reads a real field instead of guessing at text.

Fan-out is capped at 100 items and loops at 25 iterations, so a single request cannot run away.

Keep exploring