Nora — Your AI Leverage, Compounding

Nora is the AI work intelligence layer for developers. It runs silently alongside your AI coding tools, captures every session, analyzes it, and feeds that knowledge back into your next session. Every session makes the next one smarter.

Your AI Leverage Score — a composite metric of prompt quality, context injection effectiveness, decision acceptance rate, and pattern accumulation — starts at 1.0x and compounds toward 5.0x as Nora learns your patterns.

No cloud required. Run a local Ollama for fully-local analysis with no API key, or BYOK (Anthropic / OpenAI / Google / Bedrock / Grok). Your data stays on your machine.

Install

Kiro / VS Code / Cursor

Download the latest .vsix from Releases, then:

Extensions → Install from VSIX → select kernora-*.vsix

Nora bootstraps automatically — creates a Python venv, installs deps, starts the dashboard.

Claude Code

curl -fsSL https://raw.githubusercontent.com/kernora-ai/nora/main/install.sh | bash

The installer registers Nora's MCP server automatically. If you want to add it to an existing Claude Code install manually:

claude mcp add nora ~/.kernora/venv/bin/python3 -- ~/.kernora/app/nora_mcp.py

Claude Desktop (chat.claude.ai app)

The installer detects Claude Desktop and writes to ~/Library/Application Support/Claude/claude_desktop_config.json automatically. For manual setup:

{
  "mcpServers": {
    "nora": {
      "command": "/Users/YOUR_USERNAME/.kernora/venv/bin/python3",
      "args": ["/Users/YOUR_USERNAME/.kernora/app/nora_mcp.py"]
    }
  }
}

Restart Claude Desktop after editing the config.

Cowork (Claude desktop app)

Install the nora.plugin from the Releases page. After Nora is installed locally, the plugin connects Cowork to the open-core MCP tools.

What Happens

Session 1 — You code normally. Nora captures the transcript.

Between sessions — Nora analyzes the transcript: extracts patterns, architectural decisions, bugs, anti-patterns, and your AI Leverage Score.

Session 2 — Your second session starts smarter. Nora injects relevant context from past sessions directly into your prompt.

Dashboard

Open http://localhost:2742 to see:

Tab	What it shows
Home	AI Leverage Score, loop health, top projects, rule suggestions
Projects	Per-project AI metrics, patterns, decisions, bugs
Activity	Session history with outcome indicators
Coach	AI Leverage sparkline, coaching notes, decision patterns, certificate export
Knowledge	Best practices, playbooks, anti-patterns
Memory	Context injection feed, steering file viewer
Decisions	Searchable architectural decisions
Bugs	Bug inventory with severity, fix suggestions, mark resolved
Settings	LLM provider config, local AI status, telemetry

AI Leverage Score

A composite metric measuring your AI effectiveness:

AI Leverage = 1.0 + (composite_quality × 4.0)

composite_quality = (prompt_quality × 0.4)
                  + (injection_hit_rate × 0.3)
                  + (decision_acceptance_rate × 0.2)
                  + (pattern_accumulation_rate × 0.1)

Score	Label	What it means
1.0–2.0	Early	AI isn't helping much yet
2.0–3.0	Developing	Getting value, room to grow
3.0–4.0	Strong	Measurably effective AI usage
4.0–5.0	Excellent	Elite AI collaboration

Export your score as a shareable certificate from the Coach tab.

How It Works

Your IDE (Kiro / Claude Code / Cursor)
    │
    ├── hooks (capture at the moment of decision)
    │     ├── on session end → capture transcript
    │     ├── on prompt → inject relevant past context
    │     ├── on tool use → track patterns
    │     └── on session start → check steering freshness
    │
    ├── MCP server (open core — 13 tools)
    │     └── nora_search, nora_patterns, nora_decisions, nora_bugs,
    │         nora_stats, nora_session, nora_scope_validation,
    │         nora_skills, nora_retro, nora_inventory, nora_coach,
    │         nora_onboard, nora_help
    │
    ├── LLM provider (BYOK or Ollama)
    │     └── Anthropic, OpenAI, Google, Bedrock, Grok, or local Ollama
    │
    └── dashboard (localhost:2742)
          ├── Prompt-quality signals + trend
          ├── Project-level intelligence
          ├── Decision trace analysis
          ├── Loop health monitoring
          └── Steering file management

All data in ~/.kernora/echo.db. Zero bytes leave your machine in BYOK mode.

LLM Provider Priority

Nora tries these in order — the first available one wins:

1. IDE LLM (Kiro, Cursor, VS Code) — uses your IDE's built-in model, zero config
2. BYOK API keys — Anthropic, OpenAI, Google, Bedrock, Grok
3. Ollama — local, free

Use Ollama for fully-local inference.

Configuration

Edit ~/.kernora/config.toml:

[mode]
type = "byok"           # your data stays on your machine

[model]
provider = "auto"       # tries IDE → local → BYOK → Ollama

[dashboard]
port = 2742

MCP Tools

13 tools available to your AI agent in the open-core build:

Tool	What it does
nora_search	Full-text search across patterns, decisions, bugs
nora_patterns	List effective coding patterns
nora_decisions	List architectural decisions
nora_bugs	List known bugs with fixes
nora_stats	Dashboard stats
nora_session	Session details by ID
nora_scope_validation	Safety check before multi-file edits
nora_skills	Distilled team methodology
nora_retro	Engineering retrospective with git velocity
nora_inventory	Feature audit: SHIP/POLISH/WIRE/BLOCKER
nora_coach	Prompt-quality signals (sessions analyzed, avg quality, repetitions)
nora_onboard	Onboard a new developer
nora_help	List all tools

Privacy

• BYOK mode: analysis runs on YOUR machine with YOUR API key. Zero bytes reach Kernora servers.
• Ollama mode: analysis runs entirely on your machine via Ollama. No network calls.
• Team mode (coming soon): data syncs to YOUR S3 bucket. Kernora reads via a revocable IAM role you control.

Verify with tcpdump during install — the install script includes a network audit.

License

Elastic License 2.0 — free for personal and team self-hosted use. Commercial managed service requires agreement. hello@kernora.ai

Links

• Dashboard: http://localhost:2742
• Website: kernora.ai
• GitHub: github.com/kernora-ai/nora