First run of omp

Trama runs every agent job with omp. The first time, three things have to be in place: omp itself, a login to a model provider, and, optionally, the models for each kind of work. All three are in the omp and Models groups of Settings, and in the same steps of the setup wizard (see Put Trama in a project).

Install omp

When omp is not installed, Settings says so, and any agent job shows Trama: omp is not installed. Trama runs every agent with omp. Press Install omp: a terminal opens and installs omp with its official installer.

Where omp is not on the editor’s PATH, set its location with Omp Path; left empty, Trama looks for omp on the PATH, then in ~/.bun/bin and ~/.local/bin. Trama updates omp before it launches it, with a short notification Updating omp to <version>…; if the update fails, the agent runs with the version already installed.

Log in to a provider

The box at the top of the omp group says omp <version> answered with <N> models. Under Providers, press Log in beside your provider: a terminal opens and logs omp in, and the provider then shows Connected.

The omp group of Settings: a box saying omp answered with 3 models, and the Providers list where ChatGPT and Anthropic are Connected and GitHub Copilot, Google Gemini CLI and others have a Log in button.

A connected provider shows Log in again and Log out. A provider that is still connected through an environment variable or models.yml has to be disconnected there.

Choose the models

The Models group has one field for each kind of work: Analysis for design and import, Scan for scans and Execution for implementing Tasks. Leave a field empty to keep omp's default model, or pick one from the list.

The Models group of Settings with three empty model fields, Analysis described as your most capable model, Scan as a fast, light model and Execution as a mid-range model, followed by Max Parallel Agents set to 20.

Setting	Used for	Advice
Analysis	Design requests, Resolve Drifts, import	Your most capable model
Execution	Implementing Tasks, in a single Run and in Run all	A mid-range model
Scan	Scans and the sub-agents that compare Machines	A fast, light model

Where omp declares thinking levels for the model, a small selector beside the field chooses one. All settings are described in Settings.

Decide whether omp asks first

By default Trama's jobs open as conversations in omp Chat, and omp runs every command without asking (omp Chat Approval is yolo). To have omp ask permission, as it does in the terminal, choose ask.

The omp Chat and Notifications groups of Settings: the omp Chat checkbox, the omp Chat Approval selector set to yolo, omp Chat Intent Tracing, and the ntfy Server and ntfy Topic fields with Set access token and Send test notification.

The language of the agent

Language makes the agent write Tasks, Drifts, Proposals, questions and replies in another language, for example Italian. The Model, the Requirements, the Specs and the code stay in English. Empty keeps English.

Instruction files

If the project has agent instruction files (AGENTS.md, CLAUDE.md, .omp/RULES.md), the omp box lists them: the agent loads them in every Trama job, so they may conflict with Trama. Remove them from the repository, or turn on Ignore Instruction Files to make Trama’s jobs skip them.