A self-improving, always-on agent from Nous Research with a built-in closed learning loop: agent-curated persistent memory, autonomous skill creation after complex tasks, skills that self-improve during use, FTS5 cross-session search, and a Honcho dialectic user model. Runs on a $5 VPS, a GPU cluster, or serverless (Modal/Daytona).
- Storage
- Persistent memory as MEMORY.md and USER.md entries on the local filesystem; sessions indexed with SQLite FTS5 for full-text search; user-created skills persisted under ~/.hermes/skills/. Honcho provides the dialectic user-modeling layer.
- Retrieval
- FTS5 session search with LLM summarization for cross-session recall; the agent periodically nudges itself to persist knowledge and searches its own past conversations. Skills act as procedural memory and are invoked during use.
- Self-host
- Self-host: moderate
- License
- MIT
- Pricing
- Open source (MIT), free to self-host; use any model provider (Nous Portal, OpenRouter, OpenAI, own endpoint); infrastructure cost depends on chosen backend · Free / OSS
- GitHub stars
- 204,802
- Last release
- 2026-06-19
- Last commit
- 2026-06-28
- First catalogued
- 2026-06-28
Strengths
- Memory is a first-class subsystem of a complete runtime, not a bolt-on (procedural memory via skills, episodic via session search, user model via Honcho)
- Cross-platform reach: Telegram, Discord, Slack, WhatsApp, Signal, CLI from one gateway; six terminal backends incl. serverless persistence
- Very large community (200k+ stars), active release cadence (v2026.6.19, June 2026)
- Model-agnostic, no lock-in
Watch out
- Hermes is a full agent runtime, not a pure memory layer — reviewer should decide whether it belongs in a catalog of memory frameworks vs. agent harnesses (it embeds Honcho, already catalogued separately)
- Memory subsystem details (FTS5 schema, consolidation, eviction policy) are spread across the docs site, not the repo README — verify before promoting
- Self-improvement and skill-creation claims are vendor-described; not independently reproduced
Best for
- Operators wanting an always-on, self-improving personal agent whose memory (episodic + procedural + user model) is core to the loop
Benchmark results
No sourced results yet.
Sources
- https://github.com/NousResearch/hermes-agent (vendor)
- Hermes Agent docs — Memory feature (vendor)
- GitHub API repo metadata (stars, release, license) (third-party)
Last verified 2026-06-28 · updated by discover-frameworks