Ability to preview the output without updating PR, to try out different models and configurations

[lephuongbg]

It would be very helpful if CLI has a dry run mode that doesn't update the PR, but output a .md file to preview the results.

If this is possible, it will help to iterate through different models, instructions, best practices, etc. and evaluate their results quickly without spamming comment on the PR page.

[mrT23]

yep, it is supported

https://qodo-merge-docs.qodo.ai/ai_search/

[davidsantiagobillin]

How could I pass this parameter when using the service from Docker?
I tried this way, but it doesn’t work.

docker run --rm -it \ -e PUBLISH_OUTPUT=false \

[sharoneyal]

💡 Auto-generated documentation-based answer:

To run PR-Agent dockerized on your local machine without publishing comments to the PR, you need to set the publish_output parameter to false. Here's how to do it with the Docker command:

docker run --rm -it -e OPENAI.KEY=<your key> -e GITHUB.USER_TOKEN=<your token> -e CONFIG.PUBLISH_OUTPUT=false codiumai/pr-agent:latest --pr_url <pr_url> review

This will run the review tool locally without publishing any comments to the PR. The same approach works for other tools like /ask, /improve, etc.

You can also increase the verbosity level to see more output locally:

docker run --rm -it -e OPENAI.KEY=<your key> -e GITHUB.USER_TOKEN=<your token> -e CONFIG.PUBLISH_OUTPUT=false -e CONFIG.VERBOSITY_LEVEL=2 codiumai/pr-agent:latest --pr_url <pr_url> review

This configuration is useful for debugging or experimenting with different tools without affecting the actual PR.

Relevant Sources:

• https://github.com/qodo-ai/pr-agent/blob/main/docs/docs/usage-guide/automations_and_usage.md#local-repo-cli
• https://github.com/qodo-ai/pr-agent/blob/main/docs/docs/installation/locally.md#using-docker-image

[IsmaelMartinez]

This is supported, and the earlier link was to hosted Qodo docs rather than to this project. For a fully offline dry run, pipe a diff into the CLI: git diff main...feature | python -m pr_agent.cli --stdin --output review.md review, or use --diff-file changes.diff. No platform token and no PR URL are needed, results go to stdout and to the file you name, and /improve renders as one markdown document instead of inline suggestions. There is also --json-output for the review command if you want to diff runs programmatically while comparing models. If you would rather run against a real PR without posting, set publish_output=false and verbosity_level=2 under [config]. See the plain diff mode page for the full flag list.