Reco
reco.cam
- Free plan
- Open source
- Windows
- macOS
- Linux
- API
- Self-hosted
What Reco is
Reco is a desktop project that combines two camera feeds into a panoramic sports view and uses AI to follow play. Its tracking can detect balls and players, with configurable framing presets. Automatic calibration matches visual features, while IMU data or audio cross-correlation can synchronize timing between cameras. The same engine is offered as a desktop app, command-line interface, library, and OBS plugin. Reco supports GStreamer camera input and frame submission for OBS and live streams. Listed targets include Linux, macOS, Windows, NVIDIA Jetson Orin, and cloud workers. The project says it runs on hardware users own and that footage stays theirs. It also says it collects no usage data, sends no telemetry, and makes no network requests except those users configure. The engine is licensed under AGPL-3.0-only, and the project states there is no public paid tier. Coaches, parents, hardware builders, and developers are among its intended users. iOS and Android support points are present, but concrete implementations are still to come.
Who it suits
Reco may suit coaches and parents who want a panoramic view of sports play, as well as hardware builders and developers embedding its engine in their tools. Its desktop app is listed for Linux, macOS, and Windows.
Good to know, and what to watch
- Tracks detected balls and players
- Provides app, CLI, library, and OBS plugin
- Supports GStreamer input for live workflows
- States that footage stays on users’ hardware
- No public paid tier
- iOS and Android implementations are still to come
- Engine uses the AGPL-3.0-only license
Plans
Every plan and priceFree and open source
Free plan
The verdict
Reco brings camera stitching, play tracking, and live-production options together in one project. Its mobile implementations are not yet available, and teams should review the engine’s license.
What the maker publishes
- AI tracking
- AI tracking detects balls and players and applies automatic camera movement with lookahead smoothing. source
- API and extensibility
- The Rust engine exposes StitchCore frame-submission methods and reusable library crates for applications and plugins. source
- App and tools
- The project offers a desktop app, CLI, library, and OBS plugin built on the same engine. source
- Architecture
- The repository is split into nine Rust crates covering core rendering, I/O, detection, autocamera logic, calibration, controls, CLI, GUI and OBS. source
- Build requirements
- Building requires Rust 1.92 or newer, FFmpeg development libraries, pkg-config and Clang. source
- Calibration
- Automatic calibration uses AKAZE feature matching, IMU or audio cross-correlation for synchronization, and Gyroflow lens profiles. source
- Camera calibration
- Automatic calibration uses AKAZE feature matching and IMU or audio cross-correlation for temporal sync. source
- Command line interface
- The CLI supports calibration, stitching, preview, camera capture, analysis and information commands. source
- Commercial licensing
- The repository says commercial licensing questions should be directed to the maintainer and that no public paid tier exists. source
- Community size
- The GitHub repository showed 56 stars, 19 forks and five watchers when opened. source
- Company identity
- The public repository identifies the project under the reco-project GitHub organization and does not publish a legal company name, headquarters or founding year. source
- Desktop application
- Reco provides a Slint desktop application with GPU preview and export controls. source
- Detection backends
- Detection supports ONNX Runtime on CPU, CUDA and CoreML, NCNN, and native TensorRT engines. source
- Documentation
- Documentation consists of README quick-start commands, build instructions, feature flags, generated Rust-doc guidance, contributing rules and a community forum. source
- GPU support
- The renderer uses wgpu with Vulkan, Metal and DirectX 12 backends and is described as achieving more than 500 frames per second on modern desktop GPUs. source
- Hardware
- The project lists Linux, macOS, and Windows desktops, NVIDIA Jetson Orin, and cloud workers as supported targets. source
- Hardware decode
- Zero-copy decoding is available through NVDEC on Linux and Windows and VideoToolbox on macOS. source
- Intended users
- The website presents Reco for coaches and parents, hardware builders, and developers embedding the engine in their own tools. source
- Latest release
- Release v0.5.4 was published on August 7, 2026, with a macOS FFmpeg 9 compatibility build and fixes for audio alignment, model validation, checksums and vulnerable dependencies. source
- Licence
- The code is released under AGPL-3.0-only. source
- License
- The engine is licensed under AGPL-3.0-only; the repository says there is no public paid tier. source
- Limit: calibration failures
- An open issue reports repeated auto-calibration failure with two Android phones because lens matching produced too few usable feature pairs. source
- Limit: export compatibility
- The issue tracker lists HEVC exports that cannot play in QuickTime or Safari because of an hvc1 versus hev1 container tag problem. source
- Limit: image quality
- The issue tracker includes reports of pixelated or blurry exports even when the preview appears correct. source
Checked against the maker's pages on 15 Oct 2026.