Skip to main content
This guide runs TrueCourse on your own machine in local mode. You connect a repository, add its documentation, and follow the first flows to a result.

​
Before you start

  • Node.js 22 or later, pnpm 9, and Docker.
  • An API key for Anthropic, OpenAI, AWS Bedrock, GitHub Copilot or Google Gemini.
  • A project on this machine whose documentation describes what the program does.
1

Start Postgres and the server

Before you start the server, set TRUECOURSE_SECRET_KEY in .env to a strong secret of at least 32 characters. The DATABASE_URL in .env.example already matches the Postgres that docker compose starts.Open http://localhost:3000. Local mode has no sign-in. See Installation for every variable.
2

Save a model provider

Open Settings › Models. Pick a provider, enter a model and an API key, and press Test & save. The server makes a live call to the provider before it stores anything. See LLM transport.
3

Connect a repository

Open Code and press Connect repository. Pick Local folder and enter the folder’s full path on this machine. Every run works on a copy of the folder, so nothing is written into it. Go through the Context step and press Connect and start onboarding. Flow setup starts in the background.
4

Add its documentation

Open Context and press Add context. Pick Repository, choose the folder you connected, and adjust the include and exclude patterns if you need to. Press Check to count the files the scope would keep, then Add and sync. A Repository source is read by the repository it scopes, so there is nothing to link. See Sources.
5

Let the work chain

The sync starts a Document scan. When the scan changes the corpus and no conflict is open, TrueCourse starts Flow generation for the repository, and Flow generation starts a Flow run. You do not have to press anything.Agent shows each piece of work as a conversation. The repository’s Pipeline tab shows the last outcome of Flow setup, Flow generation and Flow run, each with a Run or Re-run button.
6

Read the results

Open Flows. Each flow wears one of five statuses: Succeeded, Failed, Blocked, Not testable or Never run. Open a flow to see its milestones, their cases and its tests. The repository’s Runs tab lists every run with its result and evidence.
If the scan finds a conflict between two documents, Flow generation waits until you settle it on Context › Conflicts. See Resolving conflicts.

​
Next steps

How it works

Claims, flows, milestones and cases, and how the jobs chain.

External services

Unblock flows that need an account for a third party.