All plugins

telnyx-provider

Vendor-maintained Telnyx inference provider with an authenticated live model catalog, hosted open models, frontier routes, and catalog-derived pricing.

team-telnyx1Modelsv0.1.0Updated 1 month ago

Telnyx provider for Hermes Agent

Vendor-maintained Hermes Agent
model-provider plugin for
Telnyx AI inference.

  • OpenAI-compatible chat completions at https://api.telnyx.com/v2/ai/openai
    (streaming, tool calling, usage on the final chunk).
  • Live model catalog from the authenticated /models endpoint, filtered to
    text-generation models: open-weight models hosted on Telnyx GPU
    infrastructure (Kimi, GLM, MiniMax, Qwen, Llama) plus proxied frontier
    routes.
  • Curated offline fallback models when the live catalog is unreachable.

Maintained by Telnyx. This package intentionally lives outside the
hermes-agent tree and registers through the documented external
model-provider plugin surface (providers.register_provider).

Compatibility

  • Hermes Agent >=0.21.4 (Python 3.11–3.13). Hermes 0.21.1 added flat
    provider discovery and switching, but 0.21.4 is the first release that
    correctly normalizes the Telnyx catalog's per-million-token prices.
  • CI guards the 0.21.1 host-capability floor and the minimum fully
    supported release (0.21.4) from exact release commits.

Install

hermes plugins install team-telnyx/hermes-inference-provider

The installer prompts for TELNYX_API_KEY (masked) and saves it to
$HERMES_HOME/.env (default: ~/.hermes/.env). Hermes discovers the
provider directly from the installer's flat plugin directory; no manual
move is required.

Note: the installer's "Enable now?" prompt does not gate model providers —
activation is picking the provider (hermes model or --provider telnyx).

Until the catalog entry is merged, install from the GitHub repository as
shown above. Afterwards the catalog name is enough:

hermes plugins install telnyx-provider

Create an API key at https://portal.telnyx.com/#/app/api-keys.

Restart any long-running hermes gateway after installing — provider
discovery runs once per process.

Use

hermes model                                      # interactive picker → Telnyx
hermes chat --provider telnyx -m moonshotai/Kimi-K3
hermes chat -q "hello" --provider telnyx -m zai-org/GLM-5.2

Inside an interactive session:

/model moonshotai/Kimi-K3 --provider telnyx

Models

The model list comes from the authenticated live catalog
(GET /v2/ai/openai/models), filtered to text-generation tasks. When the
live fetch fails, Hermes falls back to this curated, tool-call-verified
list:

  • moonshotai/Kimi-K3
  • moonshotai/Kimi-K2.6
  • zai-org/GLM-5.2
  • MiniMaxAI/MiniMax-M3-MXFP8

Current catalog and pricing: Telnyx model docs.

Known constraints

  • No default output cap. Combining max_tokens /
    max_completion_tokens with function tools trips Telnyx error 10015 on
    hosted models, and no catalog metadata predicts which models are
    affected. The profile therefore never volunteers a cap; only an explicit
    user agent.max_tokens sends one, and it may still be rejected by some
    models.
  • /models requires auth. Anonymous requests get a 401; without a key
    the picker shows the curated fallback list.
  • Catalog pricing is per-1M-token (pricing.unit == "1M_tokens").
    Supported Hermes releases normalize the current catalog rates. A
    provider-neutral upstream change is tracking explicit recognition of this
    unit spelling instead of relying on rate-magnitude inference.

Troubleshooting

| Symptom | Fix |
| --- | --- |
| 401 from the API | Check TELNYX_API_KEY in ~/.hermes/.env; create a key in the Telnyx portal. |
| Provider missing from hermes model | Confirm hermes plugins list shows telnyx-provider, then restart the process/gateway. Hermes 0.21.4 or newer is required. |
| Model list looks stale or short | Live catalog fetch failed (network/auth) — Hermes is showing the curated fallback list. |
| Long-running gateway doesn't see the plugin | Restart it; providers register once per process. |

Update / remove

hermes plugins update telnyx-provider
hermes plugins remove telnyx-provider

Development

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

Tests need a hermes-agent checkout on PYTHONPATH (the conftest handles
it) and nothing else beyond pytest + pyyaml; the default suite makes
no network calls.

Before a release, validate every exposed model against the authenticated
Telnyx catalog and Hermes pricing normalization:

TELNYX_LIVE_TESTS=1 HERMES_AGENT_REPO=/path/to/hermes-agent \
  python -m pytest tests/test_live_pricing.py -q

Security

Report vulnerabilities through this repository's GitHub Security
Advisories ("Report a vulnerability"). Do not open public issues for
security reports.

License

MIT

README from the reviewed commit f53c20e; it updates when the author re-pins.