Take your Codex chat to another model: aichat adds provider porting

Want to continue a Codex conversation on a local model without rebuilding its history by hand? Version 1.31.0 of the independent claude-code-tools package adds a provider option to aichat port, creating a new Codex session bound to another model provider.

The release was published November 3 at 19:27 UTC. Its linked implementation addresses a specific obstacle: according to the project’s documentation, Codex records the provider associated with a session and refuses to resume that session under a different provider. The new command makes a fork with the target provider recorded, while leaving the original session intact.

A session copy, with important context limits

The documented form is aichat port <session> --provider <name>. A user can move from OpenAI to a configured local provider, or back again. The command also accepts a model name through --model, because the old session’s last model may not exist at the destination provider.

For Codex-to-Codex transfers, the implementation preserves visible messages, tool calls, tool results and compatible history records. It creates a fresh session identifier, records its parent, writes the new session under the relevant Codex home and registers it in that home’s history. The printed resume command explicitly names the provider.

That copy is not lossless. Reasoning items and encrypted content are removed, including encrypted compaction summaries inside compacted histories. If a removed summary carried earlier context, that context is no longer visible to the destination model. The command reports dropped items; a resumed conversation should not be assumed to remember everything the source model could see.

The maintainer reports an end-to-end round trip from an OpenAI session to a local model and back, with earlier context recalled. That is a project test report, not an independent AITrending reproduction or proof of compatibility with every Codex version and provider. The reviewed release does not specify an exact minimum Codex CLI version.

Install the helper; configure model access separately

This is a community CLI utility, not an official Codex feature or a new model. The package is MIT-licensed. Its installation guide uses uv tool install claude-code-tools and requires Python 3.11 or newer; Node.js 18 or newer is listed for action menus. The separate Rust search engine is needed for search, rather than all session commands.

A non-OpenAI provider must already be defined in Codex configuration or supplied through an equivalent wrapper. Porting does not supply a local inference server, model weights, provider credentials or hosted inference credit. Inference access and costs remain those of the selected provider or the user’s own hardware.

The helper reads session transcripts and writes a new rollout plus a history entry. Resuming that copy makes the retained conversation available to the selected model provider, so provider choice also determines where that history is processed. The useful tradeoff is concrete: a conversation can travel between providers, with explicit limits on the context that can travel with it.