Skip to main content

safeailabs-provider

❖ Communityv1.0.0

Safe AI Labs as a Hermes model provider (`--provider safeailabs`, or Safe AI Labs in `hermes model`): an OpenAI-compatible API with one endpoint per region, the model list from its /v1/models and the context window of its fallback model declared.

Open in Hermes Desktop
hermes plugins install safeailabs-provider

What it adds

Environment variables it needs 1

SAFE_AI_KEY

README

From the reviewed commit 3fa82ed ↗; it updates when the author re-pins.

Safe AI Labs provider for Hermes Agent

Adds Safe AI Labs to Hermes Agent as the safeailabs model provider (alias safe-ai-labs). Safe AI Labs serves an OpenAI-compatible Chat Completions API with one endpoint per region.

Install

The plugin has no dependencies beyond Hermes 0.22.0 or later.

hermes plugins install safeailabs-provider

Before it is in the catalog, install this repository at a full commit SHA:

hermes plugins install https://github.com/nevertrust-ai/hermes-safeailabs-provider --ref FULL_COMMIT_SHA

Configure

Create a key at https://safeailabs.ai/keys and put it in ~/.hermes/.env:

SAFE_AI_KEY=sail_...

Then pick the provider:

hermes model --provider safeailabs

or a one-shot request:

hermes -z "Reply with exactly: ok" --provider safeailabs -m sail-frontier

The base URL is https://us.api.safeailabs.ai/v1. To use another region, set SAFE_AI_BASE_URL to that region's base URL, listed at https://safeailabs.ai/docs/regions.

Timeouts

Hermes drops a stream that has sent no tokens for 180 seconds on a host it does not class as local, and the API's keep-alives do not reset that clock. On a Safe AI Labs API plan a long prompt waits for the plan's token rate before its first token: about 44 minutes for a 200,000-token prompt on the smallest plan. Raise the limit for this provider in ~/.hermes/config.yaml:

providers:
  safeailabs:
    stale_timeout_seconds: 3600

Models

The model list comes from GET /v1/models on the configured base URL. When that call fails the picker offers sail-frontier, whose context window (1,048,576 tokens) the plugin declares. It declares no output limit: Hermes sends no max_tokens, and the API allows up to 65,536 output tokens.

Test

HERMES_AGENT_ROOT=/path/to/hermes-agent python -m pytest -q tests

SAFE_AI_LIVE_TEST=1 adds a check of the live /v1/models; with SAFE_AI_KEY also set it sends two billable chat requests.

Disclosure

The plugin registers one model provider at import and nothing else: no tools, hooks, commands or threads. Chat requests and the model list go to *.api.safeailabs.ai through Hermes's own client with SAFE_AI_KEY, which Hermes reads. The plugin opens no connections, reads and writes no files, starts no processes and does not update itself.

Disclosure

← Back to the catalog · catalog built Oct 20, 2026