genesisx
v0.2.2
Published
Every project begins here. Genesis reads your codebase and builds a tailored AI engineering team.
Maintainers
Readme
GenesisX
Every project begins here. GenesisX reads your entire codebase and builds a tailored team of AI agents — each owning one domain, carrying real architecture knowledge, and scoped with file-level permissions.
v0.2 — Weighted sampling, persistent memory (.genesis/state.json), dependency graph, kanban board, agent validation. 6 hardening features shipped. Agent output quality: 8-9/10.
npm install -g genesisx
cd any-project
genesis .Real Results — 5 GitHub Repositories Tested
fastapi (Python, 2935 files)
┌──────────────────────────────────────────────┐
│ fastapi · Python · 8 files │
│ DeepSeek V3 · ~$0.002 · 5 domains │
└──────────────────────────────────────────────┘
┌──────────────────────────────────────────────┐
│ files │
│ fastapi/ │
│ tests/ │
│ docs/ docs_src/ │
│ scripts/ │
│ fastapi-slim/ │
│ 8 bugs · 14 untested │
└──────────────────────────────────────────────┘
┌──────────────────────────────────────────────┐
│ team │
│ fastapi-core-engineer fastapi/ 49 files│
│ fastapi-test-engineer tests/ 581│
│ documentation-agent docs/ 2235│
│ scripts-maintainer scripts/ 69│
│ fastapi-slim-maintainer fastapi-slim/ 1│
│ reviewer all · read-only │
└──────────────────────────────────────────────┘Generated agent: fastapi-core-engineer.md (85 lines). Knows the exact FastAPI version (0.136.3), every core module, all import conventions, async patterns, test framework, and deprecation strategy — extracted straight from the source.
gin (Go, 53 files) — 2 bugs found
┌──────────────────────────────────────────────┐
│ gin · Go · 53 files │
│ DeepSeek V3 · ~$0.002 · 5 domains │
└──────────────────────────────────────────────┘
┌──────────────────────────────────────────────┐
│ files │
│ binding/ render/ internal/ ginS/ │
│ codec/ docs/ examples/ testdata/ │
│ 2 bugs · 4 untested │
└──────────────────────────────────────────────┘
┌──────────────────────────────────────────────┐
│ team │
│ gin-core-engineer core modules 54 │
│ codec-dev codec/ 5 │
│ docs-agent docs/ 1 │
│ examples-maintainer examples/ 1 │
│ testdata-domain-agent testdata/ 7 │
│ reviewer all · read-only │
└──────────────────────────────────────────────┘express (TypeScript, 10 entries) — 1 bug found
┌──────────────────────────────────────────────┐
│ express · TypeScript (Express) · 10 files │
│ DeepSeek V3 · ~$0.002 · 3 domains │
└──────────────────────────────────────────────┘
┌──────────────────────────────────────────────┐
│ files │
│ lib/ examples/ test/ │
│ 1 bug · 8 untested │
└──────────────────────────────────────────────┘
┌──────────────────────────────────────────────┐
│ team │
│ express-core-engineer lib/ 6 │
│ express-examples-agent examples/ 80 │
│ express-test-engineer test/ 112 │
│ reviewer all · read-only │
└──────────────────────────────────────────────┘click (Python, 9 entries) — clean codebase
┌──────────────────────────────────────────────┐
│ click · Python · 9 files │
│ DeepSeek V3 · ~$0.002 · 4 domains │
└──────────────────────────────────────────────┘
┌──────────────────────────────────────────────┐
│ files │
│ src/ docs/ examples/ tests/ │
│ 0 bugs · 1 untested │
└──────────────────────────────────────────────┘
┌──────────────────────────────────────────────┐
│ team │
│ core-dev src/ 18 │
│ click-documentation-agent docs/ 41 │
│ click-examples-auditor examples/ 39 │
│ click-test-engineer tests/ 31 │
│ reviewer all · read-only │
└──────────────────────────────────────────────┘got (TypeScript, 9 entries) — 3 bugs + 1 secret
┌──────────────────────────────────────────────┐
│ got · TypeScript (Express) · 9 files │
│ DeepSeek V3 · ~$0.002 · 5 domains │
└──────────────────────────────────────────────┘
┌──────────────────────────────────────────────┐
│ files │
│ source/ test/ documentation/ │
│ media/ benchmark/ │
│ 3 bugs · 1 secret · 2 untested │
└──────────────────────────────────────────────┘
┌──────────────────────────────────────────────┐
│ team │
│ got-source-agent source/ 23 │
│ got-test-domain-agent test/ 49 │
│ documentation-agent documentation/ 26 │
│ media-asset-manager media/ 4 │
│ benchmark-agent benchmark/ 2 │
│ reviewer all · read-only │
└──────────────────────────────────────────────┘Quick Start
# 1. Install
npm install -g genesisx
# 2. Set your LLM key (any one works)
export DEEPSEEK_API_KEY=sk-... # recommended: cheapest, $0.002/run
export ANTHROPIC_API_KEY=sk-ant-... # Claude
export OPENAI_API_KEY=sk-... # GPT-4o
# 3. Run on any project
cd any-project
genesis .
# Or skip approval for CI/automation
genesis . --yes
genesis . --provider claude
genesis . --dry-run # preview without writing
genesis . --json # machine-readable outputWhat Gets Created
your-project/
├── prompts/
│ ├── backend-api-engineer.md # Agent owning backend/ · 36 files
│ ├── frontend-crafter.md # Agent owning frontend/ · 12 files
│ ├── db-engineer.md # Agent owning database/ · 8 files
│ └── reviewer.md # Read-only reviewer · gates all merges
├── PROJECT-CONTEXT.md # Full AI narrative of your codebase
└── opencode.json # Registered agent teamHow It Works
Phase 1: Survey
Reads directory tree, tech stack, dependencies, imports, configs, tests.
Phase 2: Pre-Analyze (local, 0 tokens, 0 cost)
Detects hardcoded secrets, AI slop, long files, circular imports,
missing error handling, untested files, dead code.
Phase 3: Domain Detection (LLM, ~$0.001)
Sends directory structure + quality flags. LLM identifies natural
domain boundaries — the seams in your codebase.
Phase 4: Deep Analysis (LLM, ~$0.005/domain)
For small/medium domains, sends every file. For large domains, uses weighted sampling
(imports + bugs + boundary proximity) to fit context while staying high-signal.
Finds bugs at file:line precision. Detects AI-generated code. Flags security issues.
Generates the agent prompt with knowledge extracted from real code.
Phase 5: Build (local, 0 tokens)
Renders agent .md files, writes PROJECT-CONTEXT.md, registers team
in opencode.json.
Approval: You review findings before anything is written. Default yes.Providers
| Provider | Model | Cost/Run* | Context | Setup |
|----------|-------|-----------|---------|-------|
| DeepSeek | V3 | $0.002 | 128K | export DEEPSEEK_API_KEY=sk-... |
| Claude | Sonnet 4 | $0.08 | 200K | export ANTHROPIC_API_KEY=sk-ant-... |
| OpenAI | GPT-4o | $0.05 | 128K | export OPENAI_API_KEY=sk-... |
*For a 120-file project. Genesis auto-detects which provider to use from your environment.
MCP Server
Genesis also runs as an MCP server — connect it to OpenCode, Claude Code, or Cursor:
{
"mcpServers": {
"genesis": {
"command": "npx",
"args": ["genesis-mcp"]
}
}
}Tools: survey_project, propose_team, build_team. Run Genesis from inside your editor — no terminal needed.
Principles
- Every file is scanned locally. Deep analysis uses weighted sampling for large domains to fit context.
- No token limits. Choose DeepSeek and it's effectively free (~500 runs per $1).
- Your code stays local. Sent directly to the API you choose. Never cached. Never stored.
- Reviewer always included. Every team gets a read-only gate. Non-negotiable.
- File-scoped permissions. Agents can only edit their domain. Real filesystem paths.
Docs
| Doc | What | |-----|------| | PRD | What, why, who, success metrics | | Architecture | System design, data flow, provider model | | Rules | How Genesis builds itself using its own agents | | Token Costs | Per-provider cost table, no-limit strategy | | Roadmap | 14-day plan, known blockers, mitigations | | Scenarios | 5 real-world use cases with terminal output | | Difficulties | What will break and how we handle it |
MIT License · GenesisX v0.2
