\n\n
![\"dmag](\"docs/images/dmag-hero.svg\")
\n\n[](#)\n[](#)\n[](https://github.com/ATUL-Labs/dmag)\n[](https://www.linkedin.com/in/pulak-ranjan/)\n[](#)\n[](LICENSE)\n[](#)\n[](#)\n[](#platform-support)\n[](#mcp-tools)\n\n[Quick Start](#quick-start) • [CLI](#cli) • [Gateway](#gateway-zero-approval-commands) • [Viewer](#viewer) • [Skills](#skills) • [Docs](#docs)\n\n
\n\n| \n\n### Cross-agent memory\n\n`.dmag/` folder stores everything an agent needs: project state, knowledge pages,\npast mistakes, patterns, design decisions, session summaries. Any agent that can\nread a file gets the full picture — no re-explaining.\n\n**Automatic context packing:** MCP initialize injects session context (WIP, recent\nmistakes, proactive memories). `read` / `symbols` attach file-scoped packs. Failed\n`guard` / `audit_code` auto-notes CRITICAL/HIGH findings into `mistakes.md`.\n\n**Crash recovery:** `wip.md` checkpoints let the next agent resume exactly where\nthe last one left off. Switch from Claude Code to Cursor mid-task? Just type\n\"continue\".\n\n | \n\n\n### Semantic code search\n\nSelf-maintaining LMDB + MiniSearch index with 5 vectorless semantic layers: Porter\nstemmer, code synonym map, co-occurrence graph, import-graph similarity, multi-turn\ncontext. Fast path skips expansion when literal hits are enough.\n\n```bash\n$ dmag search \"auth middleware\"\n src/middleware/auth.ts:42 function validateToken()\n src/routes/api.ts:18 router.use('/api', authMiddleware)\n tests/auth.test.ts:88 test('rejects expired tokens')\n 3 files, ~412 tokens. grep would return ~12,400.\n```\n\n | \n
\n\n| \n\n### 38 MCP tools\n\nSearch, symbols, refs, links, memory, recall, guard, audit, devloop, reasoning,\nand more — all accessible from any MCP-compatible IDE. Each tool description\nexplicitly says when to use it vs built-in grep/read.\n\n```bash\ndmag mcp install --all # write configs for all 21 supported IDEs\n```\n\n | \n\n\n### Zero-approval gateway\n\nAgents run 30+ dmag commands via `write_to_file` — no `run_command` approval\nneeded, no shell quoting, no PowerShell escaping. Results auto-injected as\ncontext.\n\n```bash\nwrite_to_file('.dmag/in/r.json', 'search auth')\n# Result auto-injected. Zero user interaction.\n```\n\n | \n
\n\n| \n\n### Security & testing\n\n- **`dmag guard`** — scans for exposed secrets + DB anti-patterns before every commit\n- **`dmag audit`** — headless Chrome/Edge crawls all pages, captures console + network errors\n- **`dmag test`** — API endpoint security scanner (headers, SQL, XSS, info disclosure)\n- **`dmag devloop`** — tests all endpoints with smart categorization\n\n | \n\n\n### LAN share & docs\n\n- **`dmag share`** — sync `.dmag/pages/` across machines on the same network\n- **`dmag docs`** — distill library docs into structured JSON branches for agent consumption\n- **Docs-first instruction** — agents call `docs` tool before answering API questions\n- **Version-pinned distillation** — `npm:react@18.2.0` gets its own index entry\n\n | \n
\n\n| \n\n### Live viewer dashboard\n\n`dmag serve` opens a mission-control dashboard at `localhost:4747`:\n\n- App server status (auto-detects port)\n- Agent activity timeline\n- Full-text search + API link graph\n- Schema ERD (pannable canvas)\n- API tester + dev loop\n- Knowledge pages + session summaries\n\n | \n
\n
\n\n---\n\n## Demo\n\n\n\n| \n\n### Memory & Continuity\n\n**Dual-process memory** - conscious memory (curated `.dmag/pages/`, `status.md`, `wip.md`) + subconscious memory (passive capture of every tool call to `.dmag/subconscious/`). The `dreams` command distills subconscious into conscious, like sleep.\n\n**Continuity engine** - `wip.md` checkpoints, ~80% context flush, hooks. Sessions survive compaction, crashes, and handoffs.\n\n**Project memory** - `.dmag/` folder with knowledge pages, session summaries, audit trail. Any agent that can read a file gets the full picture.\n\n**Crash recovery** - the next agent resumes exactly where the last one left off.\n\n```bash\n$ dmag status\n# → .dmag/wip.md found - resuming task\n# → Step 3 of 5: implement auth middleware\n# → Last edit: src/auth.js (2m ago)\n```\n\n | \n\n\n### Code Intelligence\n\n**dmag-index** - self-maintaining SQLite index. Zero tokens to maintain.\n\n```bash\n$ dmag search validateToken\n# → lib/auth.js:42 function validateToken()\n# → tests/auth.test.js:88 test('rejects expired')\n# 2 files, ~248 tokens\n```\n\n**dmag guard** - scans for exposed secrets + DB anti-patterns before every commit.\n\n**Browser audit** - headless Chrome/Edge crawls all pages, captures console + network errors.\n\n**API tester** - `dmag test ` runs security scan: missing headers, SQL errors, XSS, info disclosure.\n\n**Dev loop** - `dmag devloop` tests all endpoints with smart categories: OK, auth-required, not-found, server-error.\n\n | \n
\n\n| \n\n### Agent Tools\n\n**Reasoning skills** - 16 skills: brainstorming, planning, TDD, debugging, verification, code review, and more. Each with a `HARD-GATE` - written answers required before proceeding.\n\n**Stack overlays** - PHP? Agent uses Xdebug. Rust? Agent uses `dbg!` and audits `unsafe`. 5 overlay packs auto-detect at `dmag init`.\n\n**Image converter** - `dmag convert hero.svg hero.png` - SVG to PNG/WebP/ICO via headless Chrome.\n\n | \n\n\n### Infrastructure\n\n**Gateway** - 30+ commands via `write_to_file` - zero user approval, zero shell quoting.\n\n```bash\n# Agent writes this:\nwrite_to_file('.dmag/in/r.json', 'search auth')\n# Result auto-injected as context. No approval needed.\n```\n\n**LAN share** - `dmag share start/join` syncs `.dmag/pages/` across machines on the same network. Token-based auth, dedup merge.\n\n**Structured docs** - `dmag docs distill/sync-deps/map` distills library docs into JSON branches. Agents call `docs` tool before answering API questions.\n\n**Live viewer** - `dmag serve` opens a mission-control dashboard: status, task list, knowledge, schema, search, agent activity, API tester.\n\n**Zero dependencies** - pure markdown + one small Node script. No build step, no server, nothing to update.\n\n | \n
\n
\n\nOne plugin replaces the separate reasoning, style, and memory tools you'd otherwise\nstitch together - and it works identically on **21 platforms**.\n\n---\n\n![\"dmag](\"docs/images/viewer-dark.png\")
![\"dmag](\"docs/images/viewer-light.png\")
![\"dmag](\"docs/images/viewer-panels.png\")
\n\n```bash\n# Run the viewer\nnode bin/dmag.js serve\n\n# Catch secrets before they ship\nnode bin/dmag.js guard\n\n# Search without reading files\nnode bin/dmag.js search \"userTask overdue\"\n\n# Test all endpoints with smart categorization\nnode bin/dmag.js devloop\n```\n\n---\n\n## Why it saves 98% of your tokens\n\nThe usual way an agent \"understands\" a codebase is expensive: grep for a term, read\nthe whole file that matched, read three more files to find where something else is\ndefined, read every migration to piece together the database. Each read costs tokens,\nand most of a 500-line file is irrelevant to the one function you needed.\n\n**dmag replaces that pattern with a query:**\n\n```bash\ndmag search loadAgentConfig\n# → lib/indexer.js:42 function loadAgentConfig(root) {\n# → tests/indexer.test.js:88 test('loadAgentConfig returns defaults')\n# 2 files, ~248 tokens. grep would return ~4,880.\n```\n\nOn a 45K-file repo, a common query like `ocr` returns 224 tokens via dmag vs **6.7\nmillion tokens** via grep. That's not a savings - it's the difference between working\nand crashing.\n\nFull benchmarks: [docs/benchmarks.md](docs/benchmarks.md)\n\n---\n\n![\"dmag](\"docs/images/dmag-architecture.svg\")
\n\n### Two memory systems: conscious and subconscious\n\nDmag doesn't just store files — it has a **dual-process memory system**, modeled on\nhow human memory actually works:\n\n