Where is config saved?
User defaults live in ~/.waterbrother/config.json. Project-specific overrides live in
.waterbrother/config.json inside the current workspace.
Quick answers for setup and usage.
User defaults live in ~/.waterbrother/config.json. Project-specific overrides live in
.waterbrother/config.json inside the current workspace.
~/.waterbrother/sessions/*.json
Use waterbrother resume --last, waterbrother resume <id>, or interactive /resume <id>.
Yes. It asks for provider, API key when required, model, and profile unless you pass --skip-onboarding.
The current foundation includes channel config, per-service onboarding, and readiness reporting for Telegram, Discord, and Signal. Live remote-control adapters come next.
Choose your provider, create a key in that provider's console if needed, then paste it into the wizard or run waterbrother config set provider openai and waterbrother config set apiKey YOUR_PROVIDER_API_KEY.
Use --scope project. Example:
waterbrother config set provider openai --scope project then waterbrother config set model openai/gpt-4.1 --scope project.
config get and config list default to merged runtime values.
Yes. Use waterbrother vision <image-path> <prompt> or /vision in chat.
For design work, use /agent designer.
Yes, if gh is installed and authenticated. Waterbrother now prefers a dedicated GitHub repo
reader instead of brittle shell guesses when you paste GitHub repository URLs.
Yes. It can search the open web for external or current information and fetch pasted public URLs directly. Ask for source links if you want a citation-style answer.
Yes. In interactive mode, pasting a public http:// or https:// URL will auto-read
it. You can paste just the URL or add a trailing question on the same line.
Yes. Read-only file tools can inspect ~/Desktop, ~/Downloads, and
~/Documents directly. On macOS, /desktop, /downloads, and
/documents are treated as shortcuts to those home folders.
Yes. This release supports config-driven stdio MCP servers via the mcpServers config key.
Use waterbrother config set-json mcpServers ... --scope project and verify with
waterbrother mcp list or interactive /mcp.
Yes. Use /read <url> to fetch a page directly and /search <query> to
run a web search and summarize the relevant sources. Use /open <url|index> to open a result
in your browser when a site blocks CLI fetches.
Yes. Use --approval never or interactive /approval never.
Default trace mode stays focused on tool progress. Repeating heartbeat lines such as planning state updates are now
shown in /trace verbose instead.
Default receiptMode is now auto, which suppresses noisy receipt summaries for minimal
read-only turns. Use /receipt-mode verbose or
waterbrother config set receiptMode verbose if you want every receipt printed.
Yes. Use /compact manually, and tune auto-compaction with
waterbrother config set autoCompactThreshold 0.9 (set 0 to disable).
It is project memory/source-of-truth loaded into the system prompt. Create it with
/memory init and append notes with /memory add ....
Partners should not publish from personal laptops. Use the tag-driven GitHub Actions workflow in .github/workflows/publish.yml. After the repository secret NPM_TOKEN is configured, bump the package version with npm version patch (or minor/major) and push main plus tags with git push origin main --tags. CI runs npm ci, npm run check, npm run smoke, and npm publish --access public.
Public installs use npm: npm install -g @tritard/waterbrother. Source-clone installs are only for contributors working directly in the repo.
Run waterbrother update or use /update inside interactive mode. npm installs upgrade with a global npm install -g @tritard/waterbrother@latest. Source-clone installs still use git pull, npm install, and npm run check.