Use any model and any harness
Contents
PostHog Code is harness-agnostic. You can run tasks against Claude Code (the default) or Codex, and pick a specific model within each.
Picking a harness
Choose a harness when you start a task. PostHog Code ships with two adapters today:
| Harness | Default model | Best for |
|---|---|---|
| Claude Code | Claude Opus | General-purpose code generation, long-context refactors |
| Codex | GPT‑5.x | Reasoning-heavy debugging, alternative second opinion |
More harnesses are on the roadmap – let us know on Discord which one you'd like next.
Picking a model
Each harness exposes its provider's model list. Open the model picker in the task composer to switch before sending a prompt.
Both harnesses also expose a reasoning effort level, which trades latency for deeper reasoning. The available levels differ by harness:
| Harness | Effort levels |
|---|---|
| Claude Code | low, medium, high, plus extra high and max on supported models |
| Codex | low, medium, high |
Authentication
Both harnesses authenticate through PostHog Code. You don't need to manage API keys directly.