Getting Setup
Set up your organization
From first context to org-wide knowledge layer — a phased approach that starts small and scales with proof.
The four phases
Each phase builds on the last. Start anywhere, but start with one.
1
Pick the standard that gets violated most often — the one that triggers the same review comment every sprint2
Write it as a context in the web UI: name, description, imperatives, examples3
Generate an MCP token and add it to your AI tool (Cursor, Copilot, Claude Code — 5 minutes)4
Use AI to write code in that area. See the difference immediately.✓ Outcome: You now have proof it works. One context, one team, measurable improvement.
Principles for great contexts
What separates a context that transforms workflows from one that gets ignored.
Be specific, not aspirational
"Use structured error logging with correlation IDs" beats "handle errors properly." AI needs concrete instructions, not vibes.
Include examples
A good example is worth 100 words of description. Show the pattern you want — and the anti-pattern you don't.
Assign ownership
Every context needs a human who's accountable for its accuracy. Stale contexts are worse than no contexts.
Keep it atomic
One context per topic. "Error handling" and "API naming" should be separate contexts, not a mega-doc.
Test with AI
After authoring a context, ask AI to generate code in that area. If the output doesn't match, the context needs refinement.
Measure adoption
Use analytics to see which contexts are retrieved and which are ignored. Low retrieval = needs better naming or scope.