Your AI agent is guessing which files matter

Every AI coding agent starts the same way: searching the entire repo for context. It wastes tokens on irrelevant files, misses dependencies, and builds things that already exist. Then you spend the next hour undoing its work.

CodeLedger gives your agent the right files, the right constraints, and the right patterns — before it writes a single line.

Works with Claude Code · Cursor · Codex · Gemini CLI · Windsurf · any CLI-based agent
MCP integration · Multi-session ready · Zero-config hooks

Download CodeLedger

Get the latest release. We'll only email you for major updates and to ask how it's going.

No telemetry. No tracking. This form is the only data we collect.

Downloading...

Check your downloads folder. The getting-started guide is inside the zip.

What problem are we solving?

The Problem — AI coding agents waste 40–60% of their context window on irrelevant files. Every session starts cold. Institutional knowledge lives in people’s heads and disappears when they leave. There is no risk signal before a merge.

The Solution — CodeLedger is a deterministic context control plane for software development. It scores every file in a repository, selects only what the current task requires, captures outcomes, and promotes successful patterns into reusable institutional memory.

The Intelligence Layer — The Task Intelligence Engine does not start from zero. It is seeded from day one with a curated ontology pack of golden patterns — distilled from peer organizations and leading engineering teams at organizations including Google, SAP, and Salesforce. As your team uses CodeLedger, your own earned patterns layer on top, making the system progressively more tailored to your codebase, your conventions, and your standards.

The Principle — No cloud. No training pipeline. No behavior change required. Engineering management installs it once. Every developer and every AI agent benefits automatically — from collective intelligence on day one, and from your own institutional memory from day two onward.

Logs are history. Ledger is intelligence.

One lifecycle. Every task.

CodeLedger tracks every task from intent to outcome. Each phase produces evidence, not just output.

Activate
Execute
Verify
Learn
1

Activate

Scans your repo, scores every file, and delivers a focused context bundle. The Task Intelligence Engine refines vague prompts and reports ISC confidence scores.

2

Execute

Your agent works within the context bundle. Discovery Gate checks for existing implementations. Broker delivers structured context to any surface.

3

Verify

Review Intelligence detects missing validation, unguarded I/O, circular dependencies, and brittle tests. Completion Integrity Check verifies claims against the diff.

4

Learn

Session summary shows recall, precision, and context reduction. Successful patterns are promoted. Failure hotspots compound into anti-patterns.

What CodeLedger does

Not another wrapper. A full context control plane with deterministic scoring, architectural verification, and compounding intelligence.

🎯

Deterministic Context Selection

12 weighted scoring signals per file. Dependency graph, git churn, test mappings, co-commit temporal index. Same task, same repo, same bundle.

🧠

Task Intelligence Engine

Evaluates prompt clarity (ISC scoring), refines vague tasks, and reports prompt lift. "fix bug" becomes a scoped, actionable task with 50%+ ISC improvement.

🔍

Review Intelligence

Detects missing runtime validation, unguarded HTTP calls, circular dependencies, and brittle test patterns. Auto-fix available. Zero configuration.

🛡

Discovery Gate

Scans for existing implementations before you build. Verdicts: GO (new), EXTEND (overlap), or NO_GO (already exists). Prevents duplicate systems.

📊

Engineering Intelligence Dashboard

Outcome truth, agent scorecards, destabilization metrics, and value compounding. Answers: "Is AI making the codebase more stable?" with deterministic evidence.

🔗

MCP Integration

Exposes context as MCP tools for Claude Desktop, Cursor, and Windsurf. Activation enforcement ensures no context is delivered without task linkage.

📝

Insight System

Explains what happened (explain), shows recurring patterns (learnings), and recommends next actions (next). All deterministic, no LLMs.

🏗

Architecture Health

5-component health score: duplication risk, extension discipline, source-of-truth stability, override frequency, and discovery coverage. Includes intervention engine.

🚦

Shadow: Parallel Truth Evaluation

Compare legacy and candidate implementations side by side before rollout. 5 comparators, severity classification, CI-grade gate. Never trust a refactor until the new code proves itself.

🛡

Doctrine Intelligence

Detects prompts that risk creating parallel systems or duplicate truth. Progressive intervention: none, light cue, guided refinement, two-phase stop. Seeded from 5 architectural doctrine concepts.

12
Scoring signals per file
99%
Context reduction on large repos
0
Cloud dependencies
30+
CLI commands
# Install and initialize
npm install -g @codeledger/cli
cd your-project
codeledger init

# Activate — scan + score + bundle + task intelligence
codeledger activate --task "Fix auth middleware to handle expired JWT tokens"
  Bundle: 11 files, ~8508 tokens | Confidence: HIGH
  Task Intelligence
    ISC: 0.97 [##########] sufficient
    Type: auth_change (confidence: 0.50)

# Verify — architectural checks before PR
codeledger verify --task "Fix auth middleware"

# Session recap — how well did the bundle predict your changes?
codeledger session-summary
  Bundle predicted 8/9 files you changed (89% recall)
  Context: ~8.5K tokens vs ~2.1M full repo (99% reduction)

The problem with AI coding today

AI agents are powerful — but without the right context, they're expensive guessing machines.

Wrong files, wasted tokens

Your agent reads 50 files to find the 5 that matter. You pay for the other 45 in tokens, latency, and hallucinated suggestions.

Builds what already exists

Without knowing your codebase's structure, agents create duplicate systems. You discover the overlap during code review — or worse, in production.

No memory between sessions

Every session starts from zero. Patterns that worked yesterday are forgotten today. Failures repeat. The agent never learns your codebase.

Can't prove what happened

Engineering managers ask: "Is AI making us faster?" You have vibes, not evidence. No recall metrics, no outcome truth, no audit trail.

Built on the best practices that matter

CodeLedger's Task Intelligence Engine is powered by Insight Packs — a curated ontology of 95+ prompt engineering signals and 20+ coding patterns drawn from leading AI platforms and academic research.

Curated from public best-practice documentation published by:

Anthropic OpenAI AWS Bedrock Google Gemini SAP AI Launchpad GitHub Copilot Vanderbilt Prompt Patterns CO-STAR Framework SWE-bench

7 signal categories

Meta-prompting, safety, chain-of-thought expansion, few-shot, structured output, self-critique, and enterprise governance.

Repo-calibrated

After scanning your codebase, CodeLedger generates a repo-specific pack that weights signals based on your language, framework, and patterns.

Evidence, not vibes

The Engineering Intelligence Dashboard answers the question every engineering leader asks: "Is AI making the codebase more stable — and can you prove it?"

6 real-data pages

Overview, Integrity, Quality, Knowledge, Efficiency, and Value — all computed from your actual CodeLedger session data. No synthetic, no estimates.

Engineering Intelligence

Outcome truth, agent scorecards, destabilization metrics, and value compounding. Currently in proof-of-concept mode with synthetic scenarios — real data integration on the roadmap.

Enterprise auto-deploy

Enterprise tier includes automatic dashboard deployment to GitHub Pages on your organization's account — real-time data plus a synthetic welcome scenario to explore the full feature set.

Self-hosted options

codeledger serve runs a local dashboard server. codeledger dashboard build generates static HTML. Docker and Kubernetes templates included.

Start free. Tier up when your team needs more.

Individual tier includes deterministic context selection, task intelligence, review intelligence, and the full insight system — no credit card required. Team and Enterprise tiers add MCP integration, the full dashboard, architecture governance, and fleet-wide visibility.

See full tier comparison →