Skip to content

Run through CLI

Complete project setup and install a compatible Workflow. CLI-created Runs always select service execution, even when useMcpAgent is true.

Configure the service model

Add a real provider configuration to intloom.yaml, preserving the generated Workflow declarations:

llms:
  default:
    provider: openai-compatible
    model: your-model-id
    secret: ENV.MODEL_API_KEY
    baseURL: https://your-provider.example/v1

The values above are placeholders, not a working endpoint. Supply MODEL_API_KEY securely in the environment that starts the host. Put only the reference in YAML. See Configuration for model roles and provider settings.

Service execution requires llms.default, including Code-only Workflows. Credentials for referenced Agents are checked before Run creation. An already running host retains its original configuration and environment; restart it after changes.

Start and inspect

intloom start
intloom doctor --execution cli
intloom flows

Expect the desired Workflow to be loaded and available for CLI execution. Doctor checks configuration and readiness without sending a model request; it cannot prove model compatibility or response quality.

Create one Run

In an interactive terminal:

intloom flow

Choose a Workflow and enter the intent. Alternatively, replace your_flow below with the loaded name:

intloom flow your_flow --intent "Describe the result you want"

Answer questions and review confirmation text. Requesting changes continues the same Run through the Workflow's feedback route. Enter inserts a newline in multiline prompts; Tab focuses submit and Enter submits. Cancelling an unsubmitted form leaves the Run waiting.

Continue or inspect

intloom runs
intloom attach <runId>
intloom artifacts

Replace angle-bracket placeholders with actual IDs. attach continues an existing Run; flow creates another one.

For automation, supply both a name and nonempty intent:

intloom flow your_flow --intent-file intent.txt --json

JSON mode never prompts. It returns at a waiting or terminal state. A zero exit for a wait is not completion; inspect run.status. Use an interactive attach, or a client that submits the pending action's answer, to continue. See Run management.

If execution fails, inspect intloom logs <runId> and committed results before starting another Run. A timeout does not itself cancel work. Provider output truncation, invalid structured output, and Tool failures require diagnosis; see Errors.