Kiro + Nora: Get Running in 5 Minutes
Install Nora as a Kiro Power. Your AI agent starts learning from every session. Three steps, no terminal required.
Install the Kiro Power
Open Kiro. Click the Powers panel (⚡ icon in the sidebar). Click Add power from GitHub. Paste this URL:
https://github.com/kernora-ai/nora
Click Install. Kiro clones the repo, registers Nora's MCP server, creates hooks, and copies steering file templates. Done.
Set your API key
Nora needs one LLM API key to analyze your sessions. Pick one:
export ANTHROPIC_API_KEY=sk-ant-... # Recommended export GEMINI_API_KEY=AIza... # Alternative export AWS_PROFILE=default # For Bedrock
Add it to your shell profile (~/.zshrc or ~/.bashrc) so it persists. Or use local Ollama — edit ~/.nora/config.toml and set provider = "ollama".
No API key is sent to Kernora. Nora calls the LLM directly from your machine.
Code normally
That's it. Just use Kiro like you always do. When a session ends:
→ The agentStop hook captures your session transcript
→ Nora analyzes it in the background (~60 seconds)
→ Steering files update with new patterns, decisions, and anti-patterns
→ Kiro reads these automatically on your next prompt
After 2–3 sessions, Nora has enough data to make a difference. By session 5, you'll notice Kiro avoiding mistakes it made before.
What You Get
🔍 9 MCP Tools
Kiro can search patterns, decisions, bugs, and show a full dashboard — all inside Kiro. Ask "show Nora dashboard" or "have we seen this error before?"
📋 3 Steering Files
Auto-updated markdown that Kiro reads on every prompt. Patterns, decisions, and anti-patterns from your past sessions — injected silently.
🛡 3 Hooks
agentStop captures sessions. preToolUse validates tools against known dangers. postToolUse checks output for known errors.
📊 Dashboard
Type "show Nora dashboard" in Kiro for inline stats. Or open localhost:2742 for the full web view with charts and tables.
Verify It's Working
After your first Kiro session completes, check two things:
~/.kiro/steering/ — you should see nora-patterns.md, nora-decisions.md, and nora-antipatterns.md. They may still be templates after just one session. After 2–3 sessions they'll populate with real content.
If both work, Nora is running. Every session from here makes the next one smarter.
Troubleshooting
"nora_stats not found" — The MCP server didn't register. Check that ~/.nora/app/nora_mcp.py exists. If not, run git clone https://github.com/kernora-ai/nora.git /tmp/nora-install && bash /tmp/nora-install/install.sh in your terminal.
Steering files stay empty after 3+ sessions — Check your API key is set: echo $ANTHROPIC_API_KEY (or whichever provider you chose). The analysis needs a working LLM connection.
"Permission denied" on hook scripts — Run chmod +x ~/.nora/app/*.py to make the Python scripts executable.
Dashboard not loading — Run python3 ~/.nora/app/dashboard.py & to start it manually. It should auto-start, but the first Power install may need a Kiro restart.
Privacy
Nora runs 100% on your machine. Session transcripts, analysis, and steering files never leave your computer. The only network call is to your own LLM provider for session analysis. Zero telemetry. Zero cloud storage. BYOK — your key, your machine, your data.
Next Steps
After a week: Open the dashboard at localhost:2742 and browse what Nora has learned. You'll see patterns, decisions, and anti-patterns extracted from your real sessions.
Want the full story? Read Your AI Forgets Everything — Here's How to Fix That in Kiro for the architecture rationale and all 8 MCP tool descriptions.
Using Claude Code too? The Nora installer registers a Claude Code adapter that captures Claude Code sessions into the same Nora database. Both agents share the same knowledge base — see the main repo for the adapter source.
Found a bug? Open an issue at github.com/kernora-ai/nora.
Ready?
Kiro Powers panel → Add from GitHub → paste → Install
https://github.com/kernora-ai/nora