feat: Add MaxTokens option for AI model output control

[2b3pro]

What this Pull Request (PR) does

Introduce a new MaxTokens flag and configuration option to allow users to specify the maximum number of tokens to generate in AI model responses.

This option is integrated across:

• Anthropic: Uses MaxTokens for MessageNewParams.
• Gemini: Sets MaxOutputTokens in GenerateContentConfig.
• Ollama: Sets num_predict option in chat requests.
• Dryrun: Includes MaxTokens in the formatted output.

Update example configuration to include maxTokens with a descriptive comment.

[2b3pro]

@ksylvan When I go run ./cmd/generate_changelog --ai-summarize --incoming-pr 1747 I get…

Error: PR validation failed: failed to fetch PR 1747: failed to get PR 1747: GET https://api.github.com/repos/2b3pro/fabric/pulls/1747: 404 Not Found []

Am I doing something wrong?

[ksylvan]

Are you up to date with the latest sources? I fixed this exact problem recently.

[ksylvan]

I looked at your fork on github and it looks like it's trying to run github actions (which probably won't work and may mess up other things). My fork is set up with this setting:

This is what I recommend. If your fork is up to date with main then generate_changelog should be looking at the parent fork for where to see (and summarize) the PRs.

[ksylvan]

I'll look into this more carefully tomorrow, while I'm at work, Ian. @2b3pro

[2b3pro]

I'll look into this more carefully tomorrow, while I'm at work, Ian. @2b3pro

I believe, I am up to date on main. Thank you!

[ksylvan]

I believe, I am up to date on main. Thank you!

Can you show me the output of git remote -v and git branch in your clone of Fabric?

[2b3pro]

git remove -v:

origin	https://github.com/2b3pro/fabric.git (fetch)
origin	https://github.com/2b3pro/fabric.git (push)

git branch:

  feature/structured_outputs
  feature/structured_outputs_plugin_architecture
  main
* maxTokens
  schema_plugin

[ksylvan]

Huh... I think I know what's happening. You don't have the upstream parent set up.

Do this instead (using the "gh" cli tool):

~/src $ gh repo clone ksylvan/fabric kayvan-fork  

Cloning into 'kayvan-fork'...
remote: Enumerating objects: 19649, done.
remote: Counting objects: 100% (242/242), done.
remote: Compressing objects: 100% (122/122), done.
remote: Total 19649 (delta 153), reused 120 (delta 120), pack-reused 19407 (from 3)
Receiving objects: 100% (19649/19649), 212.05 MiB | 10.14 MiB/s, done.
Resolving deltas: 100% (10027/10027), done.
remote: Enumerating objects: 6, done.
remote: Counting objects: 100% (6/6), done.
remote: Compressing objects: 100% (4/4), done.
remote: Total 6 (delta 2), reused 6 (delta 2), pack-reused 0 (from 0)
Unpacking objects: 100% (6/6), 18.01 KiB | 3.60 MiB/s, done.
From github.com:danielmiessler/Fabric
 * [new branch]        main       -> upstream/main
! Repository danielmiessler/Fabric set as the default repository. To learn more about the default repository, run: gh repo set-default --help

And now go into your cloned fork directory:

 ~/src $ cd kayvan-fork 

src/kayvan-fork [main] $ gh pr list

Showing 2 of 2 open pull requests in danielmiessler/Fabric

ID     TITLE                                                       BRANCH                CREATED AT     
#1748  feat: Implement structured output with JSON schema support  2b3pro:schema_plugin  about 1 day ago
#1747  feat: Add MaxTokens option for AI model output control      2b3pro:maxTokens      about 1 day ago

Now checkout the branch:

src/kayvan-fork [main] $ gh pr checkout 1747
remote: Enumerating objects: 31, done.
remote: Counting objects: 100% (31/31), done.
remote: Compressing objects: 100% (14/14), done.
remote: Total 16 (delta 12), reused 5 (delta 1), pack-reused 0 (from 0)
Unpacking objects: 100% (16/16), 1.97 KiB | 201.00 KiB/s, done.
From github.com:danielmiessler/Fabric
 * [new ref]           refs/pull/1747/head -> maxTokens
Switched to branch 'maxTokens'

Let's verify we are in the branch:

src/kayvan-fork [maxTokens] $ git branch
  main
* maxTokens

And now we run the change log tool to generate the snippet:

src/kayvan-fork [maxTokens] $ go run ./cmd/generate_changelog --ai-summarize --incoming-pr 1747 
Commit created successfully. Please review and push manually.
Successfully created incoming changelog entry: cmd/generate_changelog/incoming/1747.txt

Now, let's see what happened with the commit log:

src/kayvan-fork [maxTokens] $ git log --oneline | head -3
e525582b chore: incoming 1747 changelog entry
c9dda9c9 feat: Add MaxTokens option for AI model output control
70f8c013 chore(release): Update version to v1.4.307

And let's look at the generated file:

src/kayvan-fork [maxTokens] $ cat cmd/generate_changelog/incoming/1747.txt
                     
### PR [#1747](https://github.com/danielmiessler/Fabric/pull/1747) by [2b3pro](https://github.com/2b3pro): feat: Add MaxTokens option for AI model output control

- Added MaxTokens configuration option to control the maximum number of tokens generated in AI model responses
- Integrated MaxTokens support across multiple AI providers including Anthropic, Gemini, Ollama, and Dryrun
- Updated example configuration files to include maxTokens parameter with descriptive comments
- Enhanced AI model output control by allowing users to specify token limits for more predictable response lengths
- Implemented provider-specific token limiting using native API parameters (MaxTokens, MaxOutputTokens, num_predict)

And you can "git push" (or rebase into one commit and push) now.

[ksylvan]

I added an issue for myself to improve the CONTRIBUTING document.

[ksylvan]

I see you got it working - Thanks for the bug report and the coding contribution, Ian. @2b3pro

[2b3pro]

I see you got it working - Thanks for the bug report and the coding contribution, Ian. @2b3pro

Actually no. I did what you said above but got an error when I tried to git push. So I just copied the file it generated and put it into my repo.

Thank you for your patience and assistance. You're most appreciated.

[ksylvan]

Oh right... The initial gh clone should be your fork, not mine.

gh repo clone 2b3pro/fabric
cd fabric

Then follow the rest of the instructions.

[2b3pro]

Ok done!

[ksylvan]

Tests are failing @2b3pro

[2b3pro]

Thanks for your patience, @ksylvan!

[ksylvan]

@2b3pro Are you going to put forward another PR for this? If so, please add the internationalization strings in the locale files too.

[2b3pro]

@ksylvan Sure. I updated for gpt-5 which apparently changed how it manages max_tokens.

[ksylvan]

Thank you for adding the localization support, Ian.

[ksylvan]

@2b3pro Can you see what it would take to also add the max tokens parameter to LM Studio?

[2b3pro]

@2b3pro Can you see what it would take to also add the max tokens parameter to LM Studio?

@ksylvan I'll take a look

[ksylvan]

I don't understand what this has to do with the maxTokens change or why this PR is adding 800+ lines to the code base.

This seems like some leftover changes from something else, Ian @2b3pro

[2b3pro]

Looks like that's from having synced with the more recent versions of fabric