Autonomous debug agent for web apps. Scans your project, runs user flows with Playwright, collects runtime evidence, diagnoses bugs with LLM, and generates patches.
v0.2.0 · 48 tests · 8 workspaces · MIT
Work in progress — This project is not feature-complete. Feel free to fork, modify, and build on top of it.
Double-click start.command
Every time. First time it installs dependencies and builds the UI. After that it starts instantly.
Browser opens automatically → http://127.0.0.1:4317
- Open Console →
sentinel ui(browser opens automatically) - Step 1: Settings → Configure your LLM provider (API key, model)
- Step 2: Add Project → Select your web project folder
- Step 3: Run → Click Run. Sentinel will:
- Auto-initialize
.sentinel/config in your project - Auto-detect and start the dev server
- Scan routes, generate user flows
- Execute flows with Playwright
- Analyze failures with LLM
- Generate bug reports + patches
- Auto-initialize
No manual commands needed. Everything happens in the UI.
For power users who prefer the terminal:
| Command | Description |
|---|---|
sentinel ui |
Launch Web Console |
sentinel run --project=/path |
Full debug pipeline |
sentinel run --demo |
3 hardcoded sample reports (no LLM key, no real bug detection) |
sentinel run --tier 0 |
Report only, no patches |
sentinel map --project=/path |
Generate FeatureMap only |
sentinel doctor |
Check environment & config |
sentinel hello |
Test LLM connectivity |
sentinel init --project=/path |
Initialize .sentinel/ in target project |
sentinel update |
Self-update Sentinel (M8.5 stub) |
Configure via the WebUI Settings page, or edit .sentinel/llm.yml directly:
default: my-provider
providers:
my-provider:
type: openai-compatible # or ollama-native (local models)
baseUrl: https://api.example.com/v1
apiKey: your-key-here
model: model-nameSupported:
- openai-compatible — OpenAI, MiniMax, DeepSeek, Groq, Together AI, etc.
- ollama-native — Local Ollama or LM Studio (free, no API key)
Mapper → Sensor → Runner → Analyst → Critic → Planner → Enhancer → Executor
│ │ │ │ │ │ │ │
▼ ▼ ▼ ▼ ▼ ▼ ▼ ▼
FeatureMap Probes Evidence Hypothesis Confirm FixOption Enriched Patch
- Mapper — Detects frameworks, routes, data models; auto-generates user flows
- Sensor — Probes runtime state (file system, network, MCP) to collect targeted evidence per request
- Runner — Auto-starts dev server, executes flows with Playwright; emits flow.passed / flow.failed with evidence
- Analyst — LLM diagnoses root cause from collected evidence
- Critic — Validates hypothesis against evidence (confirm/reject)
- Planner — Generates fix options with cost scoring
- Enhancer — Augments fix options with external knowledge when confidence is low or 3rd-party SDK / security / version-bump is involved (skippable via
--no-enhance) - Executor — Writes patches to
.sentinel/auto-patches/(never modifies source directly)
- Auto-init — First run on a new project automatically creates
.sentinel/config - Auto dev server — Detects
dev/start/servescripts and starts the server for you - Global config — LLM settings saved once apply to all future projects
- Auto cleanup — Dev server stops automatically after debug completes
Sentinel/
├── start.command ← Double-click to install + launch
├── packages/
│ ├── core/ Microkernel (Bus / Budget / Kernel)
│ ├── subagents/ 8 agents (mapper/sensor/runner/analyst/critic/planner/enhancer/executor)
│ ├── adapters/ 13 framework detectors
│ ├── providers/ LLM / Memory / Skills / MCP / Knowledge providers
│ ├── reporters/ Markdown + JSON report generators + CLI printer
│ ├── sensors/ Evidence types + dedupe helpers (sensor agent lives in subagents)
│ └── cli/ CLI entry point
├── sentinel-ui/ React 19 + Vite Web Console
├── examples/ nextjs-blog sample project
└── docs/ Architecture, config spec, roadmap
- bun (install:
curl -fsSL https://bun.sh/install | bash) - macOS / Linux
- An LLM API key (or local Ollama for free usage)
MIT