Grounding
Your agents stop guessing.
devngn reads the real instructions, rules, and MCP config on disk and hands agents a grounded picture of the machine. Fewer hallucinated paths, fewer wrong assumptions, less cleanup after the fact.
Open source · free forever · desktop app + CLI
Ground your AI.
Ground your AI.
Protect the human.
Your AI agents guess at your setup. devngn grounds them in what is really installed, then guards the developer running them.
Grounds the agents you already run
- GitHub Copilot
- Claude
- Cursor
- OpenAI
- Gemini
- Ollama
- + local models
Agents guess. You clean up the mess.
An agent that cannot see what is installed fills the gaps with assumptions. That is where the wrong turns start:
- Calls an MCP server you removed weeks ago
- Writes to a config path that never existed
- Assumes a model you never set up
- Skips the rules sitting right there in your repo
One tool. Two jobs it refuses to compromise.
Most tools help your agents or help you. devngn does both, because the code and the coder are the same system.
The dev engine
Know your AI tooling, end to end.
- Discovers your AI-bits Finds the instructions, rules, MCP servers, model prefs, and provider config scattered across every workspace on your machine.
- Keeps agents grounded Grounds coding agents in what is actually installed and configured, so they stop guessing and start working from the truth.
- Flags drift and gaps Tracks provider readiness and token budgets across OpenAI, Copilot, Claude, Gemini, and local models, and surfaces what is stale.
Human-first wellness
Protect the person at the keyboard.
- Watches for real openings Consent-gated free/busy detection finds genuine gaps in your day. Prompts arrive when you are free, never mid-focus.
- Five-second movement breaks Short, guided resets tuned to your profile and the equipment you actually have. No gym, no context switch.
- Tracked, shared, yours Streaks, goals, and team sharing make healthy habits stick. Private by design, and deletable whenever you want.
Not a demo. This is running right now.
Every figure below comes straight from the bundled registry and runtime, rendered on each request. Providers online: OpenAI, GitHub Copilot, Anthropic Claude, Google Gemini, Local OpenAI-compatible endpoint.
16 Vendors and tools tracked
5 AI runtime providers wired
10 Ecosystem patterns watched
24 Flows with full telemetry
Readiness
Every provider, one honest view.
Track auth and SDK readiness, token budgets, and model preferences across every provider you use, local models included. Know what is ready before an agent run, not after it fails halfway through.
How the wellness loop works.
Opt in once. The engine already knows your machine, so it can respect the body running it.
- 01 Opt in Accept consent and sketch a profile: baseline, intensity, and the gear within reach.
- 02 Find the gap Connect Google or Microsoft calendars (free/busy only), or push your own events.
- 03 Take the break A five-second prompt lands in VS Code or as an OS notification from the CLI daemon.
- 04 Build the streak Complete it, keep the streak, hit goals, and bring your team along. Small breaks compound.