01
Install once
Pick npm or bun. After that the command is just okf.
Guide
Run your first OKF bundle.
Install the CLI once, run the pinned sample, use search as a progressive retrieval step for agent workflows, then switch the same workflow to your own local source path.
02
Start with the pinned sample
Use the fixed repository and commit so the first run stays reproducible, then use metadata search to stage wiki context before opening the workbench.
03
Then switch to your docs
Keep the same flow and replace only the source path, profile, and the retrieval query you want agents to use.
Step 1
Install the CLI
Install the published package once. Pick npm or bun, not both. After that every command below runs as okf.
CLI
npm install
Choose this path if npm is your package manager. Run this command or the bun command below, not both.
npm i -g @fastrag/okfor bun install
Choose this instead if bun is already part of your local workflow. You only need one global install path.
bun install -g @fastrag/okfStep 2
Run the pinned GitHub sample
Use the pinned repository and commit for the first pass. Convert, validate, run metadata search as progressive retrieval, then generate and serve the workbench locally.
CLI
convert
Read mattpocock/dictionary-of-ai-coding at the pinned commit and write the first OKF bundle into ./out/okf.
okf convert mattpocock/dictionary-of-ai-coding ./out/okf --profile ./starter-profile.en.yaml --ref 251fec7ec3b08059e4203863024e6123090a54e3validate
Check the generated bundle at ./out/okf before you move on to browser output.
okf validate ./out/okfsearch
Run metadata search against ./out/okf as the lightweight retrieval step for agent workflows before opening the workbench or loading fuller concept context.
okf search --bundle ./out/okf agentviewer generate
Generate the static workbench files into ./out/workbench so the browser UI is ready to serve.
okf viewer generate --bundle ./out/okf --out ./out/workbenchviewer serve
Start a local preview server for ./out/workbench so you can inspect the workbench in the browser.
okf viewer serve --dir ./out/workbenchStep 3
Switch to your local source path
When the sample flow looks right, replace only the source path, profile, and retrieval query. The rest of the sequence stays the same.
CLI
convert
Replace the sample repository with your own local source path at ./local-docs and write the bundle to ./out/okf.
okf convert ./local-docs ./out/okf --profile ./starter-profile.en.yamlvalidate
Run the same structural check against the local bundle output before generating browser files.
okf validate ./out/okfsearch
Replace <query> with the title, path, type, description term, or tag your agents should use for progressive retrieval before loading fuller bundle context.
okf search --bundle ./out/okf <query>viewer generate
Build the browser workbench for the local bundle into ./out/workbench.
okf viewer generate --bundle ./out/okf --out ./out/workbenchviewer serve
Serve the generated local workbench the same way as the pinned sample.
okf viewer serve --dir ./out/workbench