Summit.OS

Open source autonomous coordination for the physical world. The operational platform for wildfire response, search & rescue, disaster coordination, and commercial UAV fleets.

What it does

At 14:33 on a Tuesday, a camera on a ridge detects smoke. Here's what Summit.OS does in the next 90 seconds — with no human touching a keyboard:

Fusion receives the detection (smoke, confidence 0.91, lat 34.12, lon -118.34)
Intelligence scores it CRITICAL, dispatches the nearest available UAV to SURVEY — automatically, using a trained ML model, no LLM required
Tasking creates a mission, adjusts altitude for terrain, sends waypoints to the drone over MQTT
Fabric raises an alert, starts escalating if no operator acknowledges within your configured timeout
Console shows the operator a live map with the smoke location, the drone en route, and a live HLS video feed the moment the drone arrives

The operator's job: watch, verify, decide whether to dispatch ground resources. The software handles everything before that decision.

Why Summit.OS

The coordination software that does this well — Anduril's LatticeOS, Shield AI's platform — is closed-source, defense-export-controlled, and costs millions per year. It is not available to a county fire department, a maritime SAR team, an NGO running conservation drones, or a startup building inspection UAVs.

Summit.OS is the open-source alternative. Same architecture. Built for the civilian world.

	Summit.OS	LatticeOS / Defense platforms
License	AGPL v3	Proprietary
Access	Anyone	Defense contractors
Cost	Free to self-host	$M/year
Export restrictions	None	ITAR/EAR controlled
Use cases	Civilian emergency response, commercial ops	Military
AI	Trained on your data, self-improving	Black box

Architecture

┌─────────────────────────────────────────────────────────────────────────┐
│                         Operator Console :3002                          │
│  MapLibre live map · Alert queue · Mission feed · Live video · Replay   │
└───────────────────────────┬─────────────────────────────────────────────┘
                            │  REST + WebSocket
┌───────────────────────────▼─────────────────────────────────────────────┐
│                         API Gateway :8000                               │
│              Routing · OIDC/RBAC · Rate limiting · Audit log            │
└──────────┬──────────┬────────────────┬────────────┬─────────────────────┘
           │          │                │            │
    ┌──────▼──┐ ┌─────▼──────┐ ┌──────▼────┐ ┌────▼───────┐ ┌───────────┐
    │  Fabric │ │   Fusion   │ │Intelligence│ │  Tasking   │ │ Inference │
    │  :8001  │ │   :8002    │ │   :8003   │ │   :8004    │ │   :8006   │
    │─────────│ │────────────│ │───────────│ │────────────│ │───────────│
    │WorldSt. │ │Multi-sensor│ │Rule+ML    │ │State mach. │ │ONNX model │
    │MQTT brdg│ │track fusion│ │auto-disp. │ │Asset assign│ │hot-swap   │
    │WS stream│ │Kalman EKF  │ │Risk score │ │OPA policy  │ │YOLOv8n    │
    │Mesh CRDT│ │Re-ID       │ │LLM brain* │ │Replay store│ │           │
    └────┬────┘ └─────┬──────┘ └──────┬────┘ └────┬───────┘ └───────────┘
         │            │               │            │
    ┌────▼────────────▼───────────────▼────────────▼───────────────────────┐
    │               Redis · PostgreSQL+PostGIS · MQTT (Mosquitto)          │
    │               Prometheus · Grafana · Jaeger · OPA                    │
    └───────────────────────────────────────────────────────────────────────┘

    * Optional — system operates fully without it

Key Capabilities

Multi-sensor fusion — Kalman EKF track fusion across camera, ADS-B, AIS, MAVLink, CoT/ATAK, and any custom sensor. M-of-N track confirmation. Cross-camera re-identification.

Autonomous mission dispatch — When a detection arrives, a trained ML model (GradientBoosting, ONNX, <1ms inference) decides the mission type and dispatches immediately. Trained on 87,160 real-world labeled events (NASA FIRMS, NOAA Storm Events, GBIF) plus 20,936 synthetic examples covering edge-case mission types. No LLM required. Rules-based fallback always active.

Self-improving AI — Retrain the mission planner on your own operator decisions with one command:

python packages/ml/train_mission_classifier.py --real-data postgresql://...

The more you use it, the better it gets at your specific deployment context.

Live video — HLS streaming from any RTSP source (IP cameras, drone gimbal feeds). Sub-5-second latency. Playable in any browser.

Mission replay — Every mission is recorded as a time-indexed snapshot stream. Operators can scrub back through any incident for debrief or training.

Terrain awareness — SRTM DEM integration. All waypoints automatically adjusted to maintain true AGL altitude over real terrain.

Alert escalation — Unacknowledged alerts auto-escalate via webhook + email. Configurable timeout per severity.

CoT/ATAK compatibility — Bidirectional UDP bridge. Entities visible to any ATAK device on the network.

30-minute device integration — One base class, two methods:

class MyDrone(SummitAdapter):
    async def get_telemetry(self): ...
    async def handle_command(self, cmd): ...

Warning

Security defaults are open for local development. Out of the box, authentication is disabled (OIDC_ENFORCE=false, RBAC_ENFORCE=false, API_KEY_ENFORCE=false) and PII field encryption is off. This is intentional for local dev — running docker compose up should just work without configuring an identity provider.

Before connecting Summit.OS to any real network, real hardware, or real incident data, read the Production Hardening section below. The API Gateway will print a visible warning banner at startup until you set these values.

Quick Start

Prerequisites: Docker Desktop (or Docker + Compose V2), 8 GB RAM, 10 GB disk.

# 1. Clone
git clone https://github.com/BigMT-Ai/Summit.OS.git
cd Summit.OS

# 2. Configure (defaults work for local dev)
cp .env.example .env

# 3. Start
cd infra/docker
docker compose up

That's it. After ~60 seconds:

URL	What
http://localhost:3002	Operator console
http://localhost:8000/docs	API docs (Swagger)
http://localhost:3001	Grafana dashboards (admin / admin)
http://localhost:16686	Jaeger distributed traces

Seed demo data (see the pipeline work immediately)

pip install httpx redis
python scripts/seed_demo.py

This registers five assets on the map, injects a smoke detection, a missing hiker, and a power line alert — the intelligence service scores them, auto-dispatches missions, and your console shows a live active incident within seconds.

Add --live to keep assets moving and fire new detections every 30 seconds.

Optional: enable the LLM reasoning brain

If you have Ollama installed locally, or want to run it in Docker:

# With Docker (pulls llama3.1 on first start, ~4 GB)
docker compose --profile llm up

# Then pull the model (one time)
docker compose --profile llm exec ollama ollama pull llama3.1

The system works fully without this. The LLM adds natural-language mission reasoning for complex multi-entity scenarios.

How the AI works

Summit.OS ships with two trained ML models in packages/ml/models/:

Model	What it does	Size	Latency
`mission_classifier.onnx`	Maps (class, confidence, location) → mission type	~200 KB	<1ms
`risk_scorer.onnx`	Scores observation severity (LOW → CRITICAL)	~150 KB	<1ms

Training data breakdown:

35,928 NASA FIRMS active fire detections (global, 7-day) — real-world
49,869 NOAA Storm Events (tornadoes, floods, storm surge, 2018–2023) — real-world
1,363 GBIF wildlife observations — real-world
20,936 synthetic examples covering SEARCH, INSPECT, DELIVER, ORBIT — generated

Supported mission types:

Mission	Triggers	Asset
SURVEY	Fire, smoke, flood, earthquake damage, crop anomaly	UAV or fixed-wing
MONITOR	Person, survivor, vessel, vehicle, wildlife	UAV (loiter)
SEARCH	Missing person, distress signal, mayday, overdue vessel	UAV grid pattern
PERIMETER	Hazmat spill, tornado, storm surge, intrusion, armed threat	UAV boundary
ORBIT	Suspicious drone, vessel tracking, persistent ISR	UAV (continuous orbit)
DELIVER	Aid drop, medical supply, cargo, resupply	UAV
INSPECT	Pipeline, power line, bridge, solar farm, wind turbine	UAV (close pass)

Retrain on your data:

# Download fresh public data (NASA, NOAA, GBIF)
python packages/ml/download_real_data.py --years 2020 2021 2022 2023

# Blend with your operator-approved mission history
python packages/ml/train_mission_classifier.py \
  --real-csv packages/ml/data/real_combined.csv \
  --real-data postgresql://summit:password@localhost:5432/summit_os

# New .onnx files drop in place — no redeployment needed

Monorepo Structure

apps/
  console/          # Next.js 15 operator UI — map, alerts, missions, video
  fabric/           # WorldStore, MQTT bridge, mesh replication, WebSocket
  fusion/           # Sensor fusion, Kalman tracking, re-identification, HLS
  intelligence/     # ML dispatch, risk scoring, advisory, LLM brain (optional)
  tasking/          # Mission lifecycle, assignment, terrain following, replay
  inference/        # ONNX model serving — plug in any detection model
  api-gateway/      # Routing, auth, rate limiting, audit

packages/
  ml/               # Training pipeline, data downloaders, trained models
  entities/         # Core Entity schema — the universal data type
  world/            # WorldStore — in-memory + Postgres + broadcast
  mesh/             # CRDT replication across disconnected nodes
  geo/              # DEM terrain, elevation profiles, line-of-sight
  adapters/         # Adapter registry, built-in adapters (CoT/ATAK, MAVLink)
  observability/    # OpenTelemetry tracing middleware
  security/         # mTLS, RBAC, JWT, data classification

infra/
  docker/           # docker-compose.yml, Mosquitto, Prometheus, Grafana config
  policy/           # OPA authorization policies
  proxy/            # Nginx mTLS proxy configs

Services at a glance

Service	Port
Operator Console	3002	MapLibre map, alert queue, mission feed, video overlay
API Gateway	8000	`/docs` for full API reference
Fabric	8001	WebSocket at `/ws/{org_id}` for live entity stream
Fusion	8002	`/api/v1/tracks`, `/api/v1/video/hls`, `/api/v1/elevation`
Intelligence	8003	`/agents`, `/advisories`, `/brain/status`
Tasking	8004	`/api/v1/missions`, `/api/v1/missions/{id}/replay/timeline`
Inference	8006	`/detect`, `/models` — swap ONNX models at runtime
Grafana	3001	Pre-built dashboards for all services
Jaeger	16686	Distributed traces across the full request path

Connecting hardware

Summit.OS ships with built-in adapters for:

DJI / MAVLink autopilots — via pymavlink (set TASKING_DIRECT_AUTOPILOT=true)
CoT/ATAK — bidirectional UDP, multicast at 239.2.3.1:6969
ONVIF cameras — discovery + RTSP stream registration
OpenSky Network — live ADS-B aircraft (no account needed)
AIS vessels — maritime vessel positions (simulation mode; plug in AISHub credentials for live data)
CelesTrak — satellite orbital positions

Custom hardware: subclass SummitAdapter in packages/adapters/. See examples/ for a complete template.

Configuration

Copy .env.example to .env. Defaults work out of the box for local development.

Minimum for production (change these):

POSTGRES_PASSWORD=<strong password>
FABRIC_JWT_SECRET=<64 random hex chars>
FIELD_ENCRYPTION_KEY=<openssl rand -base64 32>

To enable AI brain (optional):

OLLAMA_URL=http://localhost:11434
OLLAMA_MODEL=llama3.1

To enforce auth:

OIDC_ENFORCE=true
OIDC_ISSUER=https://your-keycloak/realms/summit
API_KEY_ENFORCE=true
RBAC_ENFORCE=true

See .env.example for the full reference.

Development

# Infrastructure only (Postgres, Redis, MQTT, Prometheus)
docker compose -f infra/docker/docker-compose.yml up -d redis postgres mqtt prometheus grafana

# Individual services (from their app directory)
cd apps/fabric && uvicorn main:app --reload --port 8001
cd apps/console && npm run dev

# Run tests (per service — see CONTRIBUTING.md for why)
python -m pytest apps/fabric/tests/
python -m pytest apps/fusion/tests/
python -m pytest apps/tasking/tests/
python -m pytest apps/intelligence/tests/

# Hot reload all Python services
UVICORN_RELOAD=true docker compose -f infra/docker/docker-compose.yml up

Database migrations:

# Fresh install — enable PostGIS, then apply all migrations
make db-setup
make db-migrate

# Check migration state
make db-status

# Roll back one revision
make db-rollback

Migration files are in apps/fabric/alembic/versions/. Every schema change needs an Alembic migration — see CONTRIBUTING.md for the workflow.

Deployment

Self-hosted (recommended start):

# Production — builds images, no volume mounts
docker compose -f infra/docker/docker-compose.yml up --build -d

Cloud: Any Kubernetes cluster. A Helm chart is available at infra/helm/summit-os/:

helm install summit ./infra/helm/summit-os \
  --set secrets.postgresPassword=$(openssl rand -hex 32) \
  --set secrets.fabricJwtSecret=$(openssl rand -hex 32) \
  --set secrets.fieldEncryptionKey=$(openssl rand -base64 32)

Air-gapped / edge: All ML inference runs locally. No external API calls required. Designed to operate on field hardware with no internet connection.

Production Hardening

Important

The steps below are required before deploying Summit.OS to any environment where it handles real incident data or real hardware.

1. Enable authentication

# In your .env
OIDC_ENFORCE=true
OIDC_ISSUER=https://your-keycloak/realms/summit
OIDC_AUDIENCE=summit-api
OIDC_JWKS_URL=https://your-keycloak/realms/summit/protocol/openid-connect/certs
RBAC_ENFORCE=true
API_KEY_ENFORCE=true

Keycloak is the recommended identity provider — infra/docker/docker-compose.keycloak.yml stands up a pre-configured instance. Any OIDC-compliant provider (Auth0, Okta, Authentik) works.

2. Set strong secrets

POSTGRES_PASSWORD=$(openssl rand -hex 32)
FABRIC_JWT_SECRET=$(openssl rand -hex 32)
FIELD_ENCRYPTION_KEY=$(openssl rand -base64 32)
GRAFANA_ADMIN_PASSWORD=$(openssl rand -hex 16)

3. Enable MQTT TLS

The default Mosquitto config uses plaintext port 1883 — fine inside Docker's internal network, not fine if your MQTT broker is reachable externally. To enable TLS:

# Generate broker certs (writes to infra/docker/certs/)
bash scripts/gen_mqtt_certs.sh

# Then in infra/docker/mosquitto.conf:
# Uncomment the TLS listener block and comment out the dev block
# Update MQTT_PORT=8883 in your .env

The TLS configuration block is included (commented out) in infra/docker/mosquitto.conf.

4. CORS origins

Update CORS_ORIGINS to your actual console domain:

CORS_ORIGINS=https://console.yourdomain.com

5. Data retention

Configure how long audit logs and mission replays are kept:

AUDIT_RETENTION_DAYS=90    # audit log entries (default: 90)

Mission replay snapshots are stored indefinitely by default. For long-running deployments, implement a periodic cleanup of old mission_snapshots rows in Postgres based on your operational and legal requirements. SAR and emergency response operators should consult their jurisdiction's incident record-keeping requirements before reducing retention.

6. mTLS (optional, high-security deployments)

Enable the Nginx mTLS proxy layer for service-to-service encryption:

docker compose --profile mtls up

Certificates go in infra/proxy/certs/. See infra/proxy/nginx.conf for the configuration.

Editions

	Community	Enterprise
Core platform	✓ Open source (AGPL v3)	✓ On-premise
ML models (base)	✓ Included	✓ Custom trained
Observability stack	✓ Self-host	✓ Self-host or managed
SLA	—	99.9% + 4h response
Auth (OIDC/RBAC)	✓ Self-configure	✓ SSO + MFA enforced
Multi-org tenancy	✓ org_id namespacing	✓ Row-level isolation + admin UI
Custom model training	DIY	✓ On your operator data
Hardware integration support	Community	✓ Dedicated
Compliance (SOC 2 / ISO 27001)	Controls built-in, self-certify	✓ Audit support
Price	Free	Contact us

Roadmap

Helm chart for Kubernetes deployment — available at infra/helm/summit-os/
SITL testing environment — docker compose -f infra/docker/docker-compose.sitl.yml up
Pre-built adapters: Skydio, Autel, Parrot — see adapters/skydio/, adapters/autel/, adapters/parrot/
USCG SAR data integration (SEARCH model improvement)
Behavior tree editor in the console (visual mission programming)
Hardware-in-the-loop (HITL) testing with real flight controllers
Reach RTK adapter (cm-precision GPS for precision landing)
Managed hosting via BigMT.ai (self-hosted is the supported path for v0.1.0)
Federated learning — improve shared base models without sharing raw data

Contributing

Issues, PRs, and adapter contributions welcome. See CONTRIBUTING.md.

If you're a fire department, SAR team, or drone operator running Summit.OS in the field — we want to hear from you. Your operational data (anonymized) is what makes the ML models better for everyone.

Who built this

Summit.OS is a BigMT.ai project. Built for civilian operators, not defense contractors.

License

AGPL v3 — free to self-host, fork, and build on. If you deploy Summit.OS as a network service, the AGPL requires you to make your source available to users. For proprietary deployments where that's not workable, see COMMERCIAL_LICENSE.md.

The trained ML models in packages/ml/models/ are custom-built and trained by BigMT.ai and released under the same license.

Name		Name	Last commit message	Last commit date
Latest commit History 101 Commits
.github		.github
adapters		adapters
apps		apps
assets/logos		assets/logos
ci/templates		ci/templates
docs		docs
examples		examples
infra		infra
missions		missions
models		models
packages		packages
scripts		scripts
tests		tests
.env.example		.env.example
.gitignore		.gitignore
BUILDING_ON_SUMMIT.md		BUILDING_ON_SUMMIT.md
CHANGELOG.md		CHANGELOG.md
CLA.md		CLA.md
CODE_OF_CONDUCT.md		CODE_OF_CONDUCT.md
COMMERCIAL_LICENSE.md		COMMERCIAL_LICENSE.md
COMMUNITY.md		COMMUNITY.md
CONTRIBUTING.md		CONTRIBUTING.md
ENTERPRISE.md		ENTERPRISE.md
EXTERNAL.md		EXTERNAL.md
HN_POST.md		HN_POST.md
LICENSE		LICENSE
Makefile		Makefile
README.md		README.md
ROADMAP.md		ROADMAP.md
SECURITY.md		SECURITY.md
VERSION		VERSION
mypy.ini		mypy.ini
package.json		package.json
pnpm-lock.yaml		pnpm-lock.yaml
pytest.ini		pytest.ini
render.yaml		render.yaml
requirements_dev.txt		requirements_dev.txt
requirements_mock.txt		requirements_mock.txt
start_mock.sh		start_mock.sh
summit_mock.py		summit_mock.py

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

Summit.OS

What it does

Why Summit.OS

Architecture

Key Capabilities

Quick Start

How the AI works

Monorepo Structure

Services at a glance

Connecting hardware

Configuration

Development

Deployment

Production Hardening

1. Enable authentication

2. Set strong secrets

3. Enable MQTT TLS

4. CORS origins

5. Data retention

6. mTLS (optional, high-security deployments)

Editions

Roadmap

Contributing

Who built this

License

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

Summit.OS

What it does

Why Summit.OS

Architecture

Key Capabilities

Quick Start

How the AI works

Monorepo Structure

Services at a glance

Connecting hardware

Configuration

Development

Deployment

Production Hardening

1. Enable authentication

2. Set strong secrets

3. Enable MQTT TLS

4. CORS origins

5. Data retention

6. mTLS (optional, high-security deployments)

Editions

Roadmap

Contributing

Who built this

License

About

Resources

License

Code of conduct

Contributing

Security policy

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages