feat(cli): add Featherless (featherless.ai) as a built-in provider

[melodysdreamj]

Summary

Adds Featherless as a first-class, built-in inference provider. Featherless serves thousands of open-source models addressed by their Hugging Face repo IDs (e.g. zai-org/GLM-5.2) through an OpenAI-compatible API, so it is wired as a standard chat_completions API-key provider — mirroring the existing HF-style provider (GMI) with a main-model-first auxiliary design (Arcee).

Today these models are only reachable through provider: custom. This makes Featherless selectable in hermes model, fallback chains, and hermes doctor, with per-model context auto-detection.

• Base URL: https://api.featherless.ai/v1
• API key: FEATHERLESS_API_KEY (optional FEATHERLESS_BASE_URL override)
• Default model: zai-org/GLM-5.2
• Aliases: featherless-ai, featherlessai

What's included

• hermes_cli/auth.py — PROVIDER_REGISTRY entry + aliases
• hermes_cli/models.py / hermes_cli/setup.py — curated catalog (default zai-org/GLM-5.2), label, aliases
• hermes_cli/providers.py — HermesOverlay (openai_chat) + aliases
• agent/model_metadata.py — provider prefixes, URL→provider mapping, and a Featherless branch in get_model_context_length
• hermes_cli/config.py — OPTIONAL_ENV_VARS entries
• hermes_cli/doctor.py — /models health-check + env hint
• hermes_cli/main.py — routes to the generic API-key model flow
• trajectory_compressor.py — base_url host detection
• plugins/model-providers/featherless/ — provider profile plugin
• apps/desktop, .env.example, cli-config.yaml.example
• Docs: providers, quickstart, environment-variables, fallback-providers, provider-runtime, configuration
• Tests: tests/hermes_cli/test_featherless_provider.py

Provider-served context length

Featherless serves each model at a provider-specific context_length exposed via /v1/models (e.g. zai-org/GLM-5.2 at 256K, not its native 1M; an 8B Llama at 32K). The new branch in get_model_context_length prefers that authoritative endpoint value over models.dev and the hardcoded native-context fallback, so token budgeting matches what the endpoint will actually accept (mirrors the existing GMI branch).

Testing

• tests/hermes_cli/test_featherless_provider.py — registry, aliases, credentials, config registry, catalog, URL/prefix mapping, overlay, doctor hint, main-model-first aux, and context-length resolution.
• Targeted suites pass: test_featherless_provider.py, test_runtime_provider_resolution.py, test_setup_model_provider.py (168 passed).
• Live-verified against the real Featherless API through the wired Hermes code paths: chat completion with zai-org/GLM-5.2, model listing (22,275 models), and per-model context auto-detection (256K / 32K).

🤖 Generated with Claude Code

[DeamonDev888]

Good addition — #62467's unified resolver includes Featherless in its generic Category C path, so this provider will automatically benefit from the credential resolution fixes.

What this PR does: adds Featherless as a built-in provider in PROVIDER_REGISTRY.

How #62467 interacts: once Featherless is registered, the unified resolver agent/auth.py::resolve_provider_credentials() handles it automatically through the generic API-key path:

1. Check pool entry base_url first
2. Fall back to FEATHERLESS_API_KEY env var
3. Fall back to model.base_url from config.yaml (when pool_url matches registry default)
4. Validate final URL via _validate_base_url_safe (SSRF prevention)

No conflict. If this PR merges first, #62467 will automatically pick up the new provider. If #62467 merges first, Featherless just needs to be added to the registry and the resolver handles the rest.

[teknium1]

Thanks for the thorough provider integration and documentation work. This automated hermes-sweeper review is closing it under the standing third-party integration policy.

• The PR's substance is a new Featherless vendor integration in the core tree, including plugins/model-providers/featherless/ and core CLI/provider wiring (ba8998dcfffa640860d455d3953d99d334f6dcc3).
• AGENTS.md:126-136 requires third-party provider integrations to ship as standalone plugins, installable under ~/.hermes/plugins/ or via a pip entry point; this is a maintenance-coupling decision, not a quality judgment.
• The existing provider loader supports user-installed model-provider plugins at $HERMES_HOME/plugins/model-providers/ (providers/__init__.py:91-99, :163-171). Please publish Featherless through that standalone-plugin path and promote it in #plugins-skills-and-skins.

---

Closed as not-planned per standing maintainer policy (in-tree-provider-integration). This is a design-direction decision, not a code-quality judgment — see the Contribution Rubric in AGENTS.md for what the project is looking for. If you believe this policy was misapplied to your change, comment here and a maintainer will take a look.