OpenVibe.AI
Imported, written with AI assistance. Imported from the OpenVibe repository README and STATUS files at the cited commits, summarised with AI assistance; originally by OpenVibers. Not yet reviewed by a person.
OpenVibe.AI is the reusable AI authority of the OpenVibe network, on 127.0.0.1:4700 and public at https://ai.openvibe.services since 2026-10-07 (unit openvibe-ai.service). It is alpha.
What it does
The README describes it as owning "providers, models, routing, prompt templates, workflows, runs, citations, cache and quotas for every product". Products run registered, versioned workflows, and every run returns a draft and evidence package (structured output plus citations and provenance) attributed to a workflow, its version, a model and a run id, never to a person. Providers include a deterministic stub, any OpenAI-compatible API, the OpenAI Responses API, Anthropic, an HTTP seam, whisper.cpp and a local llama.cpp model. Routing can fall back, with circuit breakers; quotas refuse before any provider call; a person's own provider key is encrypted at rest. A staff operator console is server-rendered and needs OpenVibe.Network SSO.
It also answers OpenAI's Chat Completions: POST /v1/chat/completions and GET /v1/models, so a client built for OpenAI's API, such as an SDK or a coding agent, points its base URL at https://ai.openvibe.services/v1 with an app's or a first-party service's Network token. Speech-to-text runs on this host: NVIDIA Parakeet TDT 0.6B v3 for English and 24 European languages, and whisper.cpp for the rest.
What it does not do yet
The README states that AI never owns publication truth: Wiki, Blog, News, Live and the rest decide what is published. Developer-app runs stay on local capacity while no tier budget governs their project, and metered models are not yet in use there. Live's VOD and clip overviews do not yet call media.analyze; speech and embedding routes still have a single upstream and answer 503 provider.unavailable when it fails.
How it fits the network
OpenVibe.Live runs with AI_SERVICE=remote, so its shared-key calls become runs here, and OpenVibe.Network's footer copy calls network.site_copy. OpenVibe.Space's builder and a space's own AI run here, and OpenVibe.Codes' harnesses use the OpenAI-compatible endpoint. AI depends on OpenVibe.Network, OpenVibe.Events for the ai.run.* outbox relay, OpenVibe.Media for recordings, and OpenVibe.Host data roles. Publishing products call AI's workflows for drafts; AI never publishes them.
Runtime and status
Node 22 and Express on PostgreSQL (ov_ai); the local model runs as openvibe-llm.service on 127.0.0.1:8090. STATUS.json records the stage as "alpha" and the service deployed. License: AGPL-3.0.
Sources
- OpenVibe.AI README (commit abde269), retrieved
Providers, models, routing, prompt templates, workflows, runs, citations, cache and quotas for every product.
- OpenVibe.AI README (commit abde269), retrieved
points its base URL at `https://ai.openvibe.services/v1`
- OpenVibe.AI README (commit abde269), retrieved
NVIDIA Parakeet TDT 0.6B v3 (`parakeet-cli`) for English and 24 European
- OpenVibe.AI README (commit abde269), retrieved
until a second upstream exists, an unavailable sole provider answers `503 provider.unavailable`
- OpenVibe.AI README (commit abde269), retrieved
on route `space.build`
Inspect these sources (retrieval times, where each was first cited, what changed between revisions)