Assistant in the Terminal
tipeline assistant prompt (or tipeline agent prompt) does what the Desktop assistant does, from a terminal or a script: it finds the workflow, fills in what it can from your request, asks for the rest, and runs it once you confirm.
tipeline assistant prompt "give me the sales report for April"
# Monthly sales report sales-report
# Month 2026-04-01 from your request
# Format needed
# Format: pdf
# Run "Monthly sales report"? [Y/n]- When several workflows fit, it lists them and asks you to pick a number. Your pick is remembered, and shared with the Desktop app.
- A workflow from a hub is installed first, after you confirm.
- When nothing fits, it shows the workflow it would draft and asks whether to draft it. The draft is saved for you to test in the Desktop builder; it isn't run.
- When you say how often ("run the random numbers every 40 seconds"), it shows the timing under Runs on its own and asks
Create this schedule for "Random Numbers"? [Y/n]instead of running it. The values you give are used for every scheduled run. For a workflow it drafts, the timing becomes the draft's proposed schedule, which you turn on when you mark the draft ready in the Desktop app. See Schedules. - A question it can answer without running anything is answered directly.
Options
| Option | Description |
|---|---|
-y, --yes | Run, create the schedule, install from a hub, or draft without asking for confirmation. Still asks for missing values. |
-p, --project <name> | Also follow this project's instructions, on top of the global ones. Its name or identifier. |
In scripts and CI
Without a terminal to ask in, it stops — and runs nothing — when a required value is missing, when it would have to ask which workflow you mean, or when it would have to install or draft something without --yes.
| Exit code | Meaning |
|---|---|
0 | The run finished, the schedule was created, the question was answered, or you chose not to run |
1 | The run failed or was canceled, a value was missing, or something went wrong (no API key, Core not running, …) |
2 | The run paused, waiting for an approval |
Configuration
By default the CLI uses what's set in the Desktop app under Settings → Assistant — it's the same Core and the same settings. Environment variables override them:
| Variable | Overrides |
|---|---|
TIPELINE_AGENT_PROVIDER | Provider: anthropic, deepseek, gemini or openai |
ANTHROPIC_API_KEY, DEEPSEEK_API_KEY, GEMINI_API_KEY, OPENAI_API_KEY | That provider's API key |
TIPELINE_AGENT_MODEL | Model name |
TIPELINE_AGENT_BASE_URL | Base URL, for an OpenAI-compatible proxy |
TIPELINE_MCP_URL | Core's MCP address (default http://127.0.0.1:4002/mcp) |
TIPELINE_MCP_TOKEN | Token for that MCP server (default: Core's own API token) |
Set the provider together with its key
An API key variable is only read for the provider named in TIPELINE_AGENT_PROVIDER, or for DeepSeek when that isn't set. To use a key from the environment, set both:
export TIPELINE_AGENT_PROVIDER=anthropic
export ANTHROPIC_API_KEY=sk-ant-...