Start here
What do you want to try?
Each path below includes prerequisites, commands, expected results, and failure behavior.
Run a workflow
Use Docker Compose to start a coordinator, its storage, an example gRPC service, and a scripted author. In the browser, start a task, inspect its accepted result, and launch the saved workflow as an asynchronous job.
The example normalizes text and writes a record. It needs no model-provider account, host JDK, or local image build. The author is scripted; this example does not demonstrate a live model writing an arbitrary workflow.
Follow the first-workflow guide. It explains the download, sign-in credential, expected service states, job status, and how to restart without deleting stored work.
To connect the library and platform paths, project data and launch an accepted workflow. The caller projects and validates a contact before the coordinator prepares the input and runs the service steps. The guide covers output checks and retries.
For data correction, the recorded-correction guide demonstrates projection, invalid-candidate repair, validation, assessment, and offline receipt verification with a deterministic provider.
Connect a gRPC service
A service workspace stores an endpoint profile and its reflected protobuf contract. Register the endpoint once, inspect its methods, and invoke them through MCP. Remote ACP can use that same coordinator and stored profile.
The service must enable gRPC reflection and be reachable from the coordinator. The guide covers container addresses, credentials, outbound connection policy, and explicit contract refresh. Registered invocation supports unary and server-streaming calls; the walkthrough demonstrates unary calls.
Connect and invoke a service. Try a valid request and a refused request, then inspect the operation result. A successful HTTP response alone does not mean the operation succeeded.
Use the protobuf toolkit
The standalone Java example copies and selects fields, projects one message into another, validates Buf-annotated messages with ProtoMolt's runtime, stores versioned schemas with compatibility checks, and generates OpenAPI. It does not start the platform.
Run the library example. It includes complete imports, dependency configuration, expected output, and invalid inputs. JDK 25 is required for this Java-library path.
Generated OpenAPI describes supported constraints. Cross-field CEL expressions remain runtime checks, so keep validation at the application boundary.
Add inputs when you need them
Kafka and database connectors are optional. The Kafka and JDBC guide provides runnable development checks and deployment references. Named-workflow requests and accepted-workflow launch intents use different contracts and topics.
Tell us where setup gets confusing
Use the walkthrough checklist to record elapsed time, missing instructions, unexpected errors, and any outside help you needed. Do not include credentials or private input data.
Implementation and test links appear in each guide. The recorded observations distinguish automated examples from human usability results.