Skip to content
Registered users
Theme

Getting started

Go from signing in to your first exported conversations using the defaults convo ships with.

Convo ships with what a first run needs: two backends, the built-in Baseline prompt set, a pool of reviewed personas and voices, and run config defaults for the models and the judge. You can get your first conversations without writing a prompt or picking a model. These two stories take you from signing in to a JSONL file.

Generate your first conversations

You've just been given an account and want to see what convo produces before you change anything.

  1. Step 1.

    Click "Registered users" at the top of this site. Type your "Email" and "Password" and click "Sign in". Convo opens on Runs.

    The Sign in page with Email and Password fields
  2. Step 2.

    Open Backends under Settings. "Ollama Cloud" and "Ollama local" are already there. Check that "Last test" says ok on the one you'll use. If it says never, open the row with its arrow and click "Test connection".

    The Backends list with Ollama Cloud and Ollama local
  3. Step 3.

    Open Prompt sets and click "Baseline". It picks a segment for every slot of the prompt, and "Preview full prompt" shows the prompt one conversation would get. Built-in sets can't be overwritten, so it's always there to start from. See Prompt sets when you want your own.

    The built-in Baseline prompt set open
  4. Step 4.

    Open Seed preview to see the conversations a run would plan: who's talking, the mood, the occasion and the length. It calls no model and costs nothing. See Seed preview.

    The Seeds preview page with a planned table
  5. Step 5.

    Open Run configs and click "New run config". On Ollama Cloud the form already picks glm-5.3 to write the conversations and deepseek-v4.1-flash to judge them. Type a "Name", pick "Baseline" under "Prompt set", set "Target count" to something small like 10, and click "Save".

    A new run config with the default models filled in
  6. Step 6.

    Open Runs, click "Start run" and pick your config. Check the cost estimate, then click "Start run". See Runs for what the estimate covers.

    The start form with a config picked and its cost estimate
  7. Step 7.

    The run page opens and counts conversations as they're accepted or rejected. When it's done, "Reject reasons" tells you why any were turned away.

    A finished run's progress bar and counters
Next storyReview and export them

Review and export them

Your first run has finished. You want to read what came out, fix anything odd, and hand the good conversations to a trainer.

  1. Step 1.

    On the run page, click "Browse conversations" to open the run's conversations on Corpus.

    The Browse conversations link on a finished run
  2. Step 2.

    Open a conversation to read it. Click a bubble to fix a line in place. See Corpus for searching and fixing in bulk.

    A conversation open in the viewer
  3. Step 3.

    To decide on each conversation in turn, open Review and step through them with the keyboard. See Review.

    A conversation card in Review with its judge scores
  4. Step 4.

    Open Exports, click "New export", pick your run under "Run" and leave "Accepted" and "Edited" ticked. Click "Create export", and click "Download" when the file is ready.

    The export dialog with a run picked

You now have a JSONL file of accepted conversations. From here, change one thing at a time: the length mix, a cheaper model, or your own prompt set.

Next pageTutorials