All plugins

belive-aihub

BeLive AI Hub inference provider with live tool-capable chat model discovery.

belivetechModelsv0.1.0Updated 26 minutes ago

BeLive AI Hub for Hermes Agent

Use BeLive AI Hub as a named inference provider in Hermes Agent. Requires Hermes 0.22.7 or later. BeLive maintains this MIT-licensed plugin; support and source are in this repository's issues. Owner contact: Hassan / BeLive Technology.

Install and configure

Install with Hermes's supported plugin installer:

hermes plugins install https://github.com/belivetech/hermes-belive-aihub --enable

Create a workspace inference key in BeLive AI Hub. Make it available as BELIVE_TOKENHUB_API_KEY through Hermes's supported secret configuration, such as the active profile's private .env file. Keep the key out of shell history, source control and logs. Do not use a console login or management key.

Run hermes model, select BeLive AI Hub, and choose a model. For explicit configuration, merge into the active profile's config.yaml:

model:
  provider: belive-aihub
  default: qwen/qwen3.8-flash
auxiliary:
  compression:
    provider: belive-aihub
  vision:
    provider: belive-aihub
    model: qwen/qwen3.8-flash

Auxiliary requests resolve the same active profile's key. Do not add api_key overrides to auxiliary blocks. Restart any background gateway after changing secrets.

Models and billing

The plugin discovers models from https://tokenhub-api.belive.sg/api/v1/models and lists entries whose catalog explicitly advertises ready Chat Completions with tools and text output. It preserves the gateway's exact author/slug IDs. Non-chat models are excluded. If discovery fails, Hermes uses the small curated model list; that list is not a guarantee that every account may access every model. A successful empty response remains an empty list.

Hermes owns context-length and other metadata lookup. The plugin's public live_context_length hook reads the same live catalog so a named provider receives the correct context window; it does not introduce a second metadata or pricing registry. Check the AI Hub Models page for current availability and prices, and Activity for actual usage and charges. Local estimates are not an invoice.

Use Hermes's existing fallback_providers settings to opt into fallback. This plugin does not enable fallback to another service, create keys, change account limits or alter other providers. Review old auxiliary/fallback overrides when migrating.

Migration and endpoints

This provider is BeLive AI Hub, not Tencent TokenHub. It does not register the tokenhub alias or read TOKENHUB_API_KEY.

To migrate an existing BeLive custom:tokenhub configuration, keep the BeLive secret under BELIVE_TOKENHUB_API_KEY and deliberately change the provider references to belive-aihub. Existing custom connections keep working without installing this plugin. Preserve Tencent's environment variables and settings if you use that service too.

Staging qualification uses an isolated profile and a staging inference key. Set base_url: https://api-stg-tokenhub.belive.sg/api/v1 under each of model, auxiliary.compression and auxiliary.vision. A main-model override alone does not redirect explicit auxiliary-provider blocks; without their own endpoint they use the production profile default. No auxiliary key override is needed. Discovery accepts only the official production and staging HTTPS bases and refuses redirects. The host controls completion transport and explicit base-URL overrides: configure only an endpoint you intend to trust with your key. Proxy endpoints are not supported by this plugin's discovery.

Update, remove and development

Use Hermes's supported plugin update/remove commands for belive-aihub. Catalog installations are pinned to the catalog's reviewed commit; a new catalog SHA requires upstream review. Disable/remove the plugin before restoring a custom-provider configuration. No key is deleted or revoked by plugin removal.

The plugin makes authenticated model-list requests and normal host-managed inference calls to BeLive. It reads only its declared key through the host's profile-aware API or a credential passed to discovery. It has no telemetry, self-updater, credential-file reader or background service.

Behavioral tests use an isolated Hermes checkout at v0.22.7, the canonical host test runner and this repository's tests/ directory. They exercise catalog filtering, failures/redirect refusal, host discovery, and real host compression/vision credential resolution across two profiles. See VALIDATION.md for recorded commands and scope before relying on a release claim.

README from the reviewed commit 5d022b5; it updates when the author re-pins.