See it in action
2 min 45 sec · From agent chaos to deterministic control.
Runs entirely in your browser — no tracking, no external requests.
The Engineering Truth Layer
for AI Coding Agents
The Context Control Plane for AI-assisted software development. Deterministic context selection, PR review intelligence, and institutional memory — so your agents do the right work, every time.
Local-first · No cloud infra · Works with Claude Code, Cursor, Windsurf, Copilot
Free tier includes 5 context bundles/day · No credit card required
The Problem
AI coding tools are powerful.
They are also flying blind.
Every major AI coding agent has the same structural weakness: no systematic context management, no institutional memory, no way to prove the codebase is improving.
The agent forgets everything
Every session starts from zero. Yesterday's architectural decisions, last sprint's refactor — gone. Your agent makes the same mistakes twice because it has no memory.
Noisy context kills quality
Agents hallucinate when given the wrong files. Too much context wastes tokens. Too little misses critical dependencies. There's no systematic way to know which files actually matter.
Mistakes repeat across the team
One engineer figures out the right pattern. Two weeks later, someone else re-discovers the same pitfall the hard way. Hard-won lessons don't travel.
Leaders can't measure AI impact
Is the codebase getting better or worse with AI assistance? No one can answer that. There's no dashboard, no trend line, no proof — just vibes and velocity estimates.
What CodeLedger Does
Four layers. One control plane.
The right files. Scored. Every time.
CodeLedger scans your repo, scores every file against your task, and produces a minimal context bundle — ranked by dependency centrality, churn, and semantic relevance. No guesswork. No token waste.
Catch what agents miss.
Review Intelligence detects missing runtime validation, HTTP calls without timeouts, boundary violations, and circular dependencies — automatically. Every PR is checked before it merges.
Decisions persist. Lessons compound.
Every accepted change, every architectural decision, every hard-won lesson is encoded into a memory ledger. The next session starts with context the agent has never had to rediscover.
Better prompts. Better outcomes.
The Intent Sufficiency Check scores your task prompt before you write a line of code. Vague prompts get refined. High-risk instructions get doctrine signals. Good work starts before activation.
The Full SDLC
Governance from first prompt to merged PR.
CodeLedger is not a linter or a code reviewer. It is the governance layer that sits across the entire software development lifecycle.
Intent Sufficiency Check
The Prompt Coach evaluates your task description before anything runs. Vague tasks are flagged. Doctrine signals surface parallel-system risk and duplicate-truth risk before a line is written.
Discovery Gate
Before you build, CodeLedger scans for existing implementations. GO, GO_EXTENSION_ONLY, or NO_GO — so engineers extend the right system instead of creating the fourth auth helper.
Deterministic Context Selection
A scored bundle of the most relevant files — ranked by dependency graph centrality, churn, recency, and co-commit affinity. The right context. Not all the context.
Pre-Tool Enforcement
Every file write is checked against the active bundle and discovery verdict. Writes outside approved insertion points are flagged. NO_GO verdicts are blocked at the tool layer.
PR Review Intelligence
Risk, drift, and evidence gaps are computed deterministically from the diff. High-risk files are named. Missing tests are flagged. A sticky PR comment gives reviewers the signal they need.
Memory + Outcome Recording
Accepted changes update the truth ledger. Session recall and precision are recorded. Patterns that repeat get promoted. The next session starts smarter than this one.
Built for Every Role
Different views. Shared source of truth.
Get the right context without asking for it.
- ✓Context bundle auto-activates on every task
- ✓Prompt Coach catches vague tasks before you waste tokens
- ✓Institutional memory surfaces past decisions automatically
- ✓Session summary shows exactly which files you actually used
Measure AI contribution with real evidence.
- ✓Engineering Intelligence Dashboard with per-agent scorecards
- ✓Risk trends and destabilization metrics by team area
- ✓Bundle recall and precision across every session
- ✓Value Compound: hours saved with traceable formulas
Enforce architecture discipline at every PR.
- ✓Discovery Gate blocks second implementations before they ship
- ✓Review Intelligence enforces boundary constraints automatically
- ✓Doctrine signals surface parallel-system risk in prompts
- ✓Architecture Health Dashboard with trend detection
Prove AI is making the codebase better.
- ✓Executive-level Architecture Health Score (A–F graded)
- ✓Destabilization metrics tied to specific agents and PRs
- ✓Merge memory rollups by developer and time window
- ✓Truth audit: completion, topology, policy, behavior
Quantify the return on AI investment.
- ✓Hours Saved formula with conservative multipliers
- ✓Token reduction metrics tied to real context budgets
- ✓Issues prevented mapped to severity and blast radius
- ✓All calculations tagged: deterministic vs advisory
Full Feature Set
30+ commands. One coherent system.
- →Deterministic file scoring (keyword, centrality, churn, recency)
- →Shadow files via co-commit affinity graph
- →Discovery Gate — detect existing implementations pre-build
- →Intent Sufficiency Check — score prompts before activation
- →Broker API — retrieval contract for mid-session queries
- →Review Intelligence — 6 invariant modules
- →PR Review Intelligence — risk, drift, evidence gaps
- →Architecture Health Dashboard — AHS score A–F
- →Intervention Engine — prioritized fix recommendations
- →Truth Audit — 5-dimension CI gate
- →Recent Truth Ledger — accepted changes with TTL decay
- →Structural Trust — wiring neighborhoods with immutable wires
- →Ontology — concept definitions and disambiguation
- →Evidence Gates — confidence-tier rules for surfacing findings
- →Transaction Spine — auditable record of every session
- →Engineering Intelligence Dashboard — agent scorecards
- →Team Ledger — shared context across sessions
- →Multi-agent orchestration — file reservation, task partitioning
- →Policy Domain Framework — 6 governance domains
- →MCP server — expose repo memory to any MCP-compatible agent
The Difference
Raw AI agent vs. CodeLedger-governed agent.
How It Works
Up and running in under two minutes.
Install once. Runs everywhere.
npm install -g @codeledger/cli && codeledger ready. Ready initializes, scans, and wires Claude Code, Cursor, Windsurf, Codex, and MCP-compatible agents. Zero manual steps after setup.
codeledger readyContext activates automatically.
Every time you describe a task, CodeLedger scores the repo and generates a ranked context bundle. The right files arrive before you write the first line.
codeledger ready --task "..."Sessions compound. Teams improve.
Accepted changes update the truth ledger. Every session's recall and precision are recorded. The system learns what your codebase needs — and your next session starts smarter.
codeledger session-summaryBenchmark Results
Numbers you can verify.
All benchmarks run against public repos. Anyone can reproduce them.
- 28.7%
- avg token reduction
- Benchmark-proven across 8 public repos
- 84%+
- bundle recall
- Across 40 tasks
- 100%
- top-5 file stability
- Critical files always surface
- $0
- cloud infra required
- Fully local-first. No telemetry.
Benchmarks run against microsoft/vscode, prisma/prisma, and 6 other public repos. Methodology at /research/context-quality
Get started today
Your agents are ready.
Is your codebase?
CodeLedger works with the tools you already use. No cloud infrastructure, no new workflows, no mandatory meetings. Install once — your context bundles activate on every task.
Free tier · No credit card · Works locally in 2 minutes