viseron
Guides Viseron self-hosted NVR deployment, camera recording, computer-vision detection, integrations, and backend extension workflows.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- viseron
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Guides Viseron self-hosted NVR deployment, camera recording, computer-vision detection, integrations, and backend extension workflows.
- disable-model-invocation
- true
02 资源文件 6
- 参考文档references/component-capability-index.md5.6 KB
- 参考文档references/repo-provenance.md2.5 KB
- 参考文档references/repo-routing-metadata.json326 B
- 参考文档references/troubleshooting.md5.5 KB
- 脚本scripts/check_viseron_imports.py4.6 KB
- 脚本scripts/extract_config_schema_summary.py3.3 KB
03 SKILL.md 正文
Viseron Repo Skill
Use this skill when a task involves Viseron, the self-hosted local NVR and AI computer-vision application, especially when the user needs to configure cameras, recording, storage, object/motion/face/license-plate detection, notifications, automations, or backend components.
First checks
- Viseron is normally operated as a containerized service with a YAML
config.yamland optionalsecrets.yamlmounted into the runtime config directory. - The Python distribution is named
viseronand the primary import package isviseron; source checkouts also rely on a top-levelmanager.pyhelper for some subprocess modules. - Treat live cameras, Docker images, PostgreSQL, MQTT brokers, notification credentials, detector services, GPU/EdgeTPU/Hailo devices, and model downloads as target-host requirements unless the user explicitly asks to verify them.
- Run bundled scripts from this skill only against user-provided config snippets or an already prepared Viseron runtime/development environment; they are preflight helpers and do not replace Viseron startup validation.
Route by task
- Install, deploy, or validate global configuration: load configuration-and-deployment for Docker/container setup,
config.yaml,secrets.yaml, webserver/auth/API, logging, safe mode, and startup validation. - Configure camera streams, NVR, recording, storage tiers, snapshots, timelapse, or live view: load camera-recording-pipeline for FFmpeg/GStreamer/go2rtc, stream URLs, FFprobe metadata, recorder/NVR behavior, retention, and storage troubleshooting.
- Choose or debug motion/object/post-processing AI components: load detection-and-ai-components for MOG2/background subtraction, MQTT external motion, YOLO, Darknet, EdgeTPU, Hailo, CodeProject.AI, DeepStack, dlib, CompreFace, labels, masks, zones, filters, and training/post-processing touchpoints.
- Automate events or control integrations: load automation-and-integrations for MQTT/Home Assistant, webhooks, Discord/Gotify/Telegram notifications, Telegram commands, ONVIF PTZ, data-stream events, and Jinja templates.
- Modify or extend Viseron source code: load developer-extension for component/domain lifecycle, config schemas, reload behavior, entities, docs generation, tests, and source-root import caveats.
Root references and scripts
- Read component capability index when deciding which component/domain owns a user request before loading a sub-skill.
- Read cross-cutting troubleshooting for install/import, optional dependency, Docker, config, service, and source-root failures that span multiple sub-skills.
- Read repo provenance when checking whether this skill is stale relative to a newer Viseron checkout.
- Run scripts/check_viseron_imports.py in a Viseron Python environment to probe core imports and optional dependency availability without contacting cameras or services.
- Run scripts/extract_config_schema_summary.py in a Viseron source/runtime environment to list available component schemas and setup hooks.
Common workflow skeletons
Review a user configuration
- Load
configuration-and-deploymentand validate YAML/secrets syntax first. - Route each top-level component to the owning sub-skill using the component capability index.
- Check cross-component dependencies: cameras must load before NVR, object/motion detectors are optional or required depending on the component, and post-processors need object-detector labels that can trigger them.
- For every external service, credential, device, model, or camera URL, separate static config review from target-host runtime verification.
Diagnose missing recordings or detections
- Load
camera-recording-pipelineto check camera connectivity, stream metadata, recorder/NVR settings, storage tiers, and live-frame availability. - Load
detection-and-ai-componentsto check whether motion/object labels, zones, masks,scan_on_motion_only,trigger_event_recording,store, and post-processor labels can produce the expected event. - Load
automation-and-integrationsonly if the event exists in Viseron but an MQTT/webhook/notification/Telegram side effect is missing.
Add or modify a backend component
- Load
developer-extensionbefore editing code. - Keep
setup_domains()hot-reload safe: register domain instances only, and avoid expensive shared initialization there. - Put component-level resources in
setup(), declareCONFIG_SCHEMA, and model domain dependencies withRequireDomainorOptionalDomain. - After code changes, select focused native tests from the relevant sub-skill's candidate list rather than running hardware, container, or external-service checks by default.
Who owns what
| User request signal | Owning route | Why |
|---|---|---|
config.yaml, secrets.yaml, startup error, safe mode, web UI, API token, auth, or logging |
configuration-and-deployment |
These are global deployment concerns and should be checked before any camera or detector tuning. |
| RTSP/MJPEG URL, FFmpeg command, FFprobe timeout, recorder, NVR, live view, snapshots, timelapse, storage tier | camera-recording-pipeline |
These concerns are tied to frame acquisition, recordings, and retained media. |
| Motion detected but no recording, label mismatch, zone/mask issue, model/device choice, EdgeTPU/Hailo/YOLO/Darknet, face or plate recognition | detection-and-ai-components |
These are AI pipeline and post-processing choices. |
| MQTT entity/topic, webhook, Telegram, Discord, Gotify, PTZ, event topic, Jinja template | automation-and-integrations |
These are side effects or control surfaces layered on top of Viseron events. |
| Setup hook, config schema, reload diff, domain dependency, entity class, docs generation, tests, or import caveat in source code | developer-extension |
These are repo-maintenance and backend-extension tasks. |
How to use the bundled scripts
scripts/check_viseron_imports.pyprints a compact import/probe report for core modules and selected optional groups. Use it when you need to know whether a Viseron environment is ready for source inspection or which optional backend is missing.scripts/extract_config_schema_summary.pysummarizes which components exposeCONFIG_SCHEMA,setup(), andsetup_domains(). Use it when choosing which component-owned sub-skill should own a new configuration branch.- The sub-skill scripts are intentionally narrow and safe. They should fail fast on malformed local input instead of starting Viseron or contacting live hardware/services.
Verification boundaries
This skill was constructed from source, docs, tests, generated component explorer config, and CPU-safe installed-package inspection. It does not claim that this construction host verified live cameras, Docker images, PostgreSQL services, MQTT brokers, notification endpoints, YOLO/Darknet inference, EdgeTPU, Hailo, GStreamer/PyGObject, or CUDA/VA-API/Jetson/Raspberry Pi acceleration. Those remain explicit target-host checks.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).