Hermes Agent v0.5.0 expands to 400+ models and adds supply-chain hardening

What shipped

Hermes Agent v0.5.0 (The Hardening Release) on March 28, 2026.

400+ models on Nous Portal

Nous Portal now routes to 400+ models through a single provider endpoint. One subscription, one integration, the full frontier model market. This is Hermes's answer to OpenRouter for subscribers who want the managed option.

Hugging Face as first-class provider

Hugging Face is now a first-class inference provider rather than a workaround. A setup wizard walks you through API key entry, a live /models endpoint probe confirms connectivity, and a curated agentic model picker maps HF models to their OpenRouter analogues so you can pick the right one for the task.

Telegram private chat topics

Telegram now supports project-based conversations with functional skill binding per topic. Run your work project in one topic, your personal assistant in another, your finance agent in a third, all inside one chat. Each topic gets its own skill context and isolated workflow.

Plugin lifecycle hooks

Four plugin hooks activated in the agent loop and CLI/gateway:

1. pre_llm_call: run before any LLM call, good for token budgeting or redaction.
2. post_llm_call: run after, good for logging or validation.
3. on_session_start: initialisation tasks when a new session begins.
4. on_session_end: cleanup when a session ends.

Supply-chain hardening

• •Removed the compromised litellm dependency.
• •Pinned all dep version ranges (no more "install what's latest" surprises).
• •Regenerated uv.lock with hashes for reproducible builds.
• •Added CI scan for supply-chain attack patterns.

Given the 2025 litellm compromise, this is a deliberate response: no external dependency whose security posture you cannot verify.

Availability

Standard upgrade path. The litellm removal is transparent (existing calls route through Hermes's native provider interfaces).