NousResearch Launches Hermes Agent v0.8: The AI That Grows With You

NousResearch released Hermes Agent v0.8.0 on April 8, 2026, with a follow-up update on April 12 — marking the project's most feature-complete release yet. Unlike conventional chat assistants that reset with every session, Hermes Agent lives on a server, accumulates memory, and synthesizes new skills from every interaction it has with users. The company's tagline captures the philosophy succinctly: "the agent that grows with you."

At the core of Hermes Agent is what NousResearch calls a "persistent learning loop." After each session, the agent reviews what it learned, writes or updates skill definitions based on novel problems it encountered, and nudges itself to preserve critical knowledge for future use. It also maintains a searchable record of past conversations, allowing it to surface relevant context from weeks or months ago — behavior that begins to resemble long-term human memory.

The v0.8 release adds several significant capabilities: background task auto-notifications, live model switching across all active platforms without service interruption, native Google AI Studio integration, MCP OAuth 2.1 support, and smart inactivity timeouts that reduce idle cloud costs. Users can interact with a single Hermes Agent instance across Telegram, Discord, Slack, WhatsApp, Signal, and email — the agent maintains a unified context regardless of which channel a message arrives on.

Model flexibility is a key differentiator. Hermes Agent is not tied to any single AI provider. Users can route inference through Nous Portal, OpenRouter (with access to over 200 models), OpenAI, Google AI Studio, or any custom endpoint. Switching models requires a single command — no code changes, no redeployment. This design insulates users from vendor lock-in and allows them to optimize cost or capability on the fly.

Deployment is intentionally lightweight. NousResearch says the agent can run on a -per-month virtual private server, a GPU cluster for heavier workloads, or a serverless setup that costs almost nothing during idle periods. The project is open-source on GitHub, and a growing community is already contributing new platform integrations and skill libraries.