FAQ
What can Codebuff be used for?
Software development: Writing features, tests, and scripts across common languages and frameworks. It can also run CLI commands, adjust build configs, review code, and answer questions about your repo.
What model does Codebuff use?
Multiple. The orchestrator ("Buffy") uses Claude Opus 4.5 in Default and Max modes, or Grok 4.1 Fast in Lite mode. Subagents are matched to their tasks: GPT-5.1 and Claude Opus 4.5 for code editing, Gemini 2.5 Pro for deep reasoning, Grok 4 Fast for terminal commands and research, and Relace AI for fast file rewrites. See What models do you use? for the full breakdown.
Is Codebuff open source?
Yes. It's Apache 2.0 at github.com/CodebuffAI/codebuff.
Do you store my data?
We don't store your codebase. The server forwards requests to model providers. We keep small slices of chat logs for debugging.
Do you use model providers that train on my codebase or chat data?
No, we don't choose providers that will train on your data in our standard modes.
Can I trust Codebuff with full access to my terminal?
If you want isolation, use the Dockerfile to run Codebuff against a scoped copy of your codebase.
Can I specify custom instructions for Codebuff?
Yes. Add knowledge.md files to describe patterns, constraints, and commands. Codebuff also reads AGENTS.md and CLAUDE.md if present. Per directory, it picks one: knowledge.md first, then AGENTS.md, then CLAUDE.md. Codebuff updates existing knowledge files but won't create them unless you ask.
Can I tell Codebuff to ignore certain files?
Codebuff by default will not read files that are specified in your .gitignore. You can also create a .codebuffignore file to specify additional files or folders to ignore.
What if I want Codebuff to not ignore some files in the .gitignore?
The .codebuffignore follows standard .gitignore negation rules as well! So to ignore a file from git but not from Codebuff:
path/to/file
!path/to/file
Take note that if you want to negate something in a nested subdirectory, you need to first negate each of the parent directories first (assuming they were ignored by some other rule). So something like:
# ignore everything**
# negate: path/to/file!path/to/!path/to/file
The final set of files ignored by Codebuff are determined by the contents of .gitignore, followed by the contents of .codebuffignore.
How does Codebuff work?
Codebuff runs specialized models in parallel: one finds files, another reasons through the problem, another writes code, another reviews. A selector picks the best output. In Max mode, multiple implementations compete. Benchmark comparison.
How does Codebuff compare to Claude Code?
See Codebuff vs Claude Code for a detailed comparison. Short version: Codebuff is faster, cheaper, and handles large codebases better.
I have more questions!
Contact support@codebuff.com or join our Discord.