Abstract Provider for Hermes Agent

This plugin adds Abstract's inference API to Hermes Agent as the abstract model provider. Hermes talks to https://api.abstract.inc/v1 over the OpenAI Responses API (codex_responses mode) with live streaming.

Install

The plugin has no runtime dependencies beyond Hermes. Install it from this repository, pinned to a full commit SHA:

hermes plugins install https://github.com/abstractpbc/hermes-abstract-provider --ref FULL_COMMIT_SHA

Replace FULL_COMMIT_SHA with the full reviewed commit hash. Review the installer's scan and consent prompts.

Configure

Get an API key at platform.abstract.inc, or with the abstract CLI (abstract login, then abstract token). Set it in the active Hermes profile's environment:

export ABSTRACT_API_KEY="your-api-key"

Select the provider and a model:

hermes model --provider abstract

Or in config.yaml:

model:
  provider: abstract
  default: glm-5.3:superfast

One-shot request:

hermes -z "Reply with exactly: ok" --provider abstract -m glm-5.3:superfast

Models

Model	Input $/M	Cached input $/M	Output $/M
glm-5.3:superfast	1.19	0.26	4.40
deepseek-v4-flash:superfast	0.10	0.05	0.25
kimi-k3:superfast	3.00	0.30	12.75

Prices are per million tokens; see abstract.inc/pricing for the current rates. Output is capped at 65,536 tokens per response. Input is text only. The API refuses request bodies over 2 MiB.

The plugin uses a static catalog of these three models: Abstract's /models endpoint also lists background (:async) lanes, which complete by submit-and-poll and are not offered here. deepseek-v4-flash:superfast is the default auxiliary model (compression, titles).

What it does

The plugin registers one model provider at import and nothing else: no tools, hooks, commands, threads or files. Hermes' own client sends chat requests (prompts, context and tool definitions) to https://api.abstract.inc/v1 with ABSTRACT_API_KEY, which Hermes reads. Requests are billed to the key's Abstract account.

License

MIT