43 MCP tools, 6 sub-agent workflows, and 190+ security rules. Knowledge graphs, blast radius analysis, and closed-loop learning across Claude Code, OpenCode, Codex, Cursor, and more.
Prefer a local install? Run npx milens init --profile full --interactive
Knowledge graphs, blast radius prediction, security scanning, and self-learning memory — all through a single MCP server.
Real-time code intelligence with automatic symbol extraction, dependency resolution, and FTS5 semantic search. 1,432 symbols and 2,386 links indexed in milens itself — updated every time you save.
Explore tools →190+ built-in rules covering OWASP Top 10. Secret detection, injection prevention, hidden Unicode, dependency auditing. Replace 10 manual grep calls with one security_scan() tool.
7 pre-built MCP prompts: Planner, Reviewer, Tester, Architect, Security Auditor, Debugger, Dead Code Remover. Your coding agent calls 1 prompt instead of chaining 5–10 tools — saving thousands of tokens per task.
Skill catalog →Three reasons developers choose Milens over raw LLM coding — and stick with it.
Milens builds a SQLite knowledge graph of your entire codebase. 1,432 symbols, 2,386 links, all queryable in milliseconds. Milens answers "who calls this?" instantly.
Agents annotate discoveries. Milens remembers. Over time, patterns evolve into permanent rules. No other tool has memory.
Milens' compact format uses 40-Efficient Token Usage than raw code blocks. For teams using AI daily, that's real budget impact.
The fastest path from zero to production-grade AI coding. One command installs everything.
npx milens init --profile full analyzes your entire codebase with tree-sitter, builds a knowledge graph, and generates everything your AI agent needs.
AGENTS.md auto-loads with codebase context. Skills are available. Security rules are active. Your agent understands the project before the first prompt.
edit_check() guards every edit. impact() predicts blast radius. review_pr() assesses risk. Your agent codes safely, every session.
The knowledge graph drives intelligence. The hook system drives automation. The platform drives distribution.
The knowledge graph, semantic search, blast radius engine, and 43 MCP tools that make agents smarter.
Hooks, auto-annotation, watch mode, and scheduled evolution — the system learns and improves itself.
GitHub App, 7 harness adapters, desktop dashboard, and npm distribution — reach every developer.
See how Milens upgrades the same workflow without forcing a tool reset.
codebase_summary() — 500 tokens, complete overviewimpact() — exact blast radius before every editsecurity_scan() — 50 rules, one tool callrecall() — agent remembers every past lesson7 pre-built sub-agent prompts. Each bundles multiple milens tools into a single, repeatable workflow.
Implementation planning with blast radius awareness. Research → Analyze → Predict Impact → Plan Tests → Final Plan.
Most UsedAutomated PR risk assessment. Scores every changed symbol CRITICAL/HIGH/MEDIUM/LOW based on heat, dependents, and test coverage.
Team FavoriteTest-driven development enforcer. Find coverage gaps → generate test plans → implement → verify. Keep tests first, coverage visible.
EssentialOWASP Top 10 audit. Secret detection, injection scanning, hidden Unicode, dependency audit — all in one prompt.
SecuritySafe refactoring with dead code detection. Find unused symbols → verify with context → check blast radius → clean up.
AdvancedRoot cause analysis with execution tracing. Context → Trace → Impact → Relationship → Root Cause.
Advanced"Milens cut our token waste by 70%. The knowledge graph means agents understand the codebase instantly instead of reading 15 files blind. The ROI was visible in the first session."
Developer, SaaS Startup
"The blast radius analysis alone saved us from 3 production incidents this month. Knowing exactly what breaks before we change a hub function is a game-changer."
Tech Lead, Fintech
"We replaced 10 manual security grep calls with one security_scan tool. 50 rules, OWASP mapped, zero false positives on our codebase. It just works."
Security Engineer
Built-in security engine covering OWASP Top 10. Secrets, injections, hidden Unicode, dangerous patterns — caught before they reach production.
AWS keys, GitHub tokens, OpenAI API keys, private keys, hardcoded passwords. 10 rules with confidence scores. 10 rules
SQL injection, XSS, command injection, eval(), exec(), dangerous DOM APIs. 9 rules covering OWASP A03:2021. 9 rules
Zero-width characters, bidi override attacks, homoglyph detection. Catches prompt injection vectors that are invisible to the human eye. 4 rules
One install path. Start free on public repos, move to Pro for private repositories and team automation.
Best for OSS maintainers and evaluation on public repos.
For teams that need private repo coverage and active-seat billing.
Free proves fit, Pro supports real teams.
OSS stays free. The core is MIT-licensed forever.
Milens is AI-DOS — the Operating System for AI-Driven Development. It builds a SQLite knowledge graph of your codebase (functions, classes, imports, calls, and inheritance chains) and exposes it through 43 MCP tools. Your AI agent queries the graph instead of reading files blind. The system learns from every session via annotations that can auto-promote to permanent rules.
Three key differences. First: knowledge graph vs grep. Tools like Copilot and Cursor search files blindly; Milens pre-indexes everything into a graph — "who calls this?" is one API call, not 5 greps and 8 file reads. Second: blast radius before every edit. Before modifying any function, Milens shows exactly what will break (depth-3 impact analysis). Third: memory across sessions. Annotations persist in SQLite. Over time, patterns auto-promote to permanent rules. No other tool has memory.
No. Zero network. Zero telemetry. Zero data leaving your machine. The knowledge graph is stored in a local .milens/ SQLite database inside your repo (gitignored). The MCP server binds to 127.0.0.1 only. Optional embeddings are generated locally via Xenova transformers — no API calls. Milens is MIT-licensed and fully auditable.
190+ built-in security rules mapped to OWASP Top 10 across 9 categories: secret detection (AWS keys, tokens, passwords), injection prevention (SQL, XSS, command), hidden Unicode attacks (zero-width chars, bidi override), dangerous patterns, insecure config, data leaks, weak crypto, auth issues, and path traversal. Offline CVE database audits dependencies across 5 ecosystems (npm, Python, Rust, Go, Java). One security_scan() call replaces 10 manual greps.
Yes — this is the core innovation. Every session, your agent calls annotate() to record discoveries (bugs, caveats, patterns). Next session, recall() retrieves them. Each recall boosts confidence (+0.05). When confidence reaches 0.8, milens evolve auto-promotes the annotation to a permanent SKILL.md file that enforces the rule for every future session. It's a real self-improving system, not just a prompt wrapper.
One command: npx milens init --profile full. This analyzes your codebase, builds a knowledge graph, generates AGENTS.md with full context, installs 13+ skill files, configures 190+ security rules, and sets up pre-commit hooks. Then connect your AI agent via MCP (7 editors supported). The built-in file watcher auto-reindexes on every save — no manual steps needed. Open your agent and start coding smarter.