Modes

The assistant works in one of three modes, selected in the chat composer. Each mode is set up for a different task.

Plan

Plan is the default. Use it when you know the outcome you want but not the exact components. Describe the goal, such as "collect nginx logs, parse them, and send them to a backend," and the assistant proposes a design and asks follow-up questions. The assistant's reasoning streams as it plans, then the plan is presented as a card headed Configuration Plan Ready with three choices. Build it hands the plan to Build. Tweak it keeps you in Plan and the assistant asks what to change; describe the change, and a revised plan card follows. Just Planning closes the card as Just planning — saved for reference; switch to Build later to build it. After any choice, the card's buttons are replaced by a read-only badge. A plan keeps the configuration's $orgvar: and $uservar: references as they are and never proposes replacing one with a literal endpoint or credential.

Plan can also propose tests for the current configuration without changing it. A test plan's card is headed Test Plan Ready, lists up to ten planned tests with a plain-language rationale for each, summarizing any beyond that as +N more tests, and its approve button is Create tests. Every planned test, including any the card summarizes, is handed to Build when you approve; Build authors and runs them and saves the assertions it ran, whether they passed or failed.

Build

Build turns a description or an instruction into configuration. Use it to generate a new pipeline, or to modify the canvas, such as adding a processor or changing an endpoint. Build applies its changes to the canvas and YAML for you to review and save, and verifies changes that affect telemetry by running a pipeline test in the sandbox, at most three per message. Assertions it checks are saved automatically to the pipeline's validation rules, marked as created by the AI assistant; if they cannot be saved, the test result still stands and the assistant tells you the assertions were not saved. Rules saved with telflo test --save or through the API are marked Created via the Telflo API; the assistant replaces only the rules it created, and the CLI and the API, which share that marker, replace only rules saved through the API; rules you author by hand are never touched. Build keeps $orgvar: and $uservar: references as written and substitutes the organization's values only into the temporary copy it validates and tests. See Organization variables.

Ask

Ask answers questions without changing the configuration. Use it to understand how a component works or to compare options. If you ask it to build or change something, it points you to Build or Plan mode instead, and if you ask it to test a pipeline, it explains what could be asserted and tells you to switch to Build.

Switch modes

Switch modes at any time in the composer. A common sequence is Plan to design, Build to generate, and Ask to explain.