# Choose agents and models

import { Steps } from '@astrojs/starlight/components';

Choose which agents answer a journey and which model each one runs. Both affect which reports you can compare over time.

## Which agents a journey can use

| Agent | Type | Research | Agent Task | Reports the sources it read |
| --- | --- | --- | --- | --- |
| **ChatGPT** | AI assistant | Yes | No | Yes |
| **Claude** | AI assistant | Yes | No | Yes |
| **Perplexity** | AI assistant | Yes | No | Yes |
| **Gemini** | AI assistant | Yes | No | Yes |
| **Claude Code** | Coding agent | Yes | Yes | Yes |
| **Codex** | Coding agent | Yes | Yes | No |

- A Research journey can use any of the six, so you can also ask a product question the way a developer would ask it inside a coding agent.
- An Agent Task journey uses coding agents only. See [Agent Task journeys](/journeys/agent-task-journeys/).
- A new journey starts with **ChatGPT**. Switching it to **Agent Task** selects **Claude Code**. At least one agent stays selected.

## Models and defaults

You select one model per agent. The menu shows each vendor's own model ID, and the first item is the default.

| Agent | Default | Other choices |
| --- | --- | --- |
| ChatGPT | `gpt-6.1-sol` | `gpt-6-astra`, `gpt-6-luna`, `gpt-5.6-sol`, `gpt-4o` |
| Claude | `claude-sonnet-5-5` | `claude-opus-5-5`, `claude-haiku-4-5-20251001`, `claude-sonnet-5` |
| Gemini | `gemini-3.8-flash` | `gemini-3.1-pro-preview`, `gemini-3.5-flash-lite`, `gemini-2.5-flash` |
| Perplexity | `sonar` | `sonar-pro`, `sonar-reasoning-pro`, `sonar-deep-research` |
| Claude Code | `claude-sonnet-5-5` | `claude-opus-5-5`, `claude-haiku-4-5-20251001` |
| Codex | `gpt-5.3-codex` | `gpt-6.1-sol`, `gpt-5.2` |

The list changes as vendors release models; the menu in the app is always current. A journey keeps the model it was saved with, even after EkLine changes a default or drops that model from the menu.

Each report records, under **Run configuration and environment**, the model each vendor said it served. A vendor can serve a newer build under the same name, which is why the app marks this "recorded, not pinned".

## Before you begin

- A journey to edit, or a new one. See [Create a research journey](/journeys/create-a-research-journey/).
- Any member of your organization can change a journey.

## Steps

<Steps>
1. Open the journey's **Edit** tab, or click **New journey**.

2. Under **How we'll run it**, click an agent's name to add or remove it.

3. To change a selected agent's model, click the model name on its chip and select one from the **Model for** menu.

4. Click **Save draft**, or run the journey.
</Steps>

## Settings that differ by agent

| Agent | Turns web search off when you clear **Let the agent search the web** | Applies the **At most** search limit |
| --- | --- | --- |
| ChatGPT, Gemini, Claude Code, Codex | Yes | No |
| Claude | Yes | Yes, as one budget for the whole run |
| Perplexity | No, it always searches | No |

Agent Task journeys do not show the web search control. Each run records the search settings that were actually applied.

## What you cannot choose

EkLine picks the model that grades judged checks, and the models behind **Create scenario with AI** and the written summaries in reports. You cannot change them. See [How results are measured](/journeys/how-results-are-measured/).

## How a change affects comparisons

- **Each agent has its own trend.** Results are never averaged across agents.
- **Changing an agent's model starts a new series** for that agent, with a new baseline. Its earlier reports stay on the **Reports** tab under their own heading, which says "deltas are comparable within this series only".
- **Adding or removing an agent** starts or ends that agent's series and leaves the other agents' series as they were.
- **Changing Number of runs or web search does not start a new series.** Reports in one series can therefore differ in those settings. See [Track results over time](/journeys/track-results/).

## Check it worked

- Under **How we'll run it**, each selected agent's chip shows the model it runs.
- After the next run, the report's **Run configuration and environment** lists each agent with its model.

## Next

- [Journey agents and checks](/reference/journey-agents-and-checks/): the reference tables.
- [Track results over time](/journeys/track-results/)