AI-sorted signal harvester for public forums. Surfaces pointers to Discord; never writes a reply.
773
One container that watches public forums for the conversations you care about, has an AI sort each hit into your buckets, scores it, and posts a short pointer into a Discord channel. A human sees the pointer, raises a hand, and replies in their own words. The AI finds the conversation; a person has it. The tool never writes a reply.
The topic is not baked in: it's config. One beacon watches observability pain, another watches people wrestling with what AI means for them, another could watch gas prices or potholes. One binary runs many at once, each just a config block. Pure open source, zero Immersive Fusion dependency: bring your own keys and run it.
sos-beacon reads secrets from the environment and a small YAML config. Grab config.example.yaml, edit it, then:
docker run --rm \
-e OAI_KEY -e OAI_BASE_URL -e SOS_APM_WEBHOOK \
-v "$PWD/config.yaml:/config.yaml:ro" \
-v "$PWD/sos-state.json:/sos-state.json" \
immersivefusion/sos-beacon -config /config.yaml
OAI_KEY / OAI_BASE_URL: any OpenAI-compatible endpoint (OpenAI, Azure OpenAI, Ollama, vLLM, OpenRouter, Groq).SOS_APM_WEBHOOK (and SOS_AI_WEBHOOK, ...): a Discord channel webhook URL per beacon.One pass, or always on. The command above runs a single pass and exits, the zero-infra path for a cron or a GitHub Actions schedule. Add -interval 10m to run it as a long-lived container that polls each beacon on its own schedule and shuts down cleanly on SIGTERM.
Persistence. The dedupe/claim ledger must survive between runs. Either mount the JSON file store shown above (sos-state.json), or point it at Azure SQL (store: { type: azuresql, ... }, or SOS_STORE_DSN) for a long-running or multi-instance deployment, no volume needed. Miss this in one-shot mode and the beacon re-posts everything.
The image is multi-arch (linux/amd64, linux/arm64), distroless, and runs as non-root.
sos-beacon surfaces pointers: a place to look, a score, a bucket, a link. It does not draft replies, write outreach, or auto-post. That's not a setting; there is no code path from the AI's verdict to a posted message. See the refused-features list.
The container is OpenTelemetry-instrumented, and every LLM classification is a span with full OTel GenAI semantic conventions (model, token usage, latency, the verdict). Immersive Fusion runs the flagship beacons as live demo grids that stream into DeepCube (TM)'s 3D player, so you can walk the real pipeline as it moves: sources polling, the model judging each post, embeds firing. Unlike a synthetic demo, this is a real workload on public display. When a source fetcher dies, you watch the phantom form in the topology.
latest follows the newest release; each release also publishes a matching semantic-version tag. Pin by digest for reproducible cluster runs.github.com/ImmersiveFusion/sos-beacon
Product docs: docs.deepcube.ai
Apache-2.0. Built by Immersive Fusion.
Content type
Image
Digest
sha256:0e46cb821…
Size
7.4 MB
Last updated
2 months ago
docker pull immersivefusion/sos-beacon