AI/TLDR

Nous Research · 2026-05-28 · major

Nous Research Ships Hermes Agent v0.15.0 'The Velocity Release' — Multi-Agent Kanban Platform, 76% Smaller Core Loop, 4,500× Faster Session Search

Velocity release of the self-improving Hermes Agent collapses run_agent.py from 16,083 to 3,821 lines, ships an orchestrator-driven Kanban swarm, and adds Promptware defense against Brainworm-class injections.

GitHub release page banner for Hermes Agent v0.15.0 The Velocity Release

Velocity release rewires Hermes Agent into a multi-agent Kanban swarm and shrinks the core loop by 76%.

Key specs

Versionv0.15.0
GitHub stars171,524
Commits1,302
Merged prs747
Files changed1,746
Issues closed560
Contributors321
Core loop loc16083 -> 3821 (-76%)
Cold start ms701 -> 258
Session search speedup4500x

What is it?

Hermes Agent is Nous Research's open-source MIT-licensed agent framework that creates skills from experience, persists memory across sessions, and runs on 200+ models via OpenRouter, OpenAI, and Anthropic. v0.15.0 lands 1,302 commits, 747 merged PRs, and 1,746 changed files from 321 contributors. The release positions the project as a swarm-capable agent platform instead of a single-loop chatbot.

How does it work?

The 16,083-line run_agent.py was decomposed across 14 cohesive agent/* modules, leaving a 3,821-line core. The Kanban subsystem gained orchestrator auto-decomposition, swarm topology, scheduled tasks, worktree-per-task isolation, and per-task model overrides across 104 PRs. session_search was rebuilt without an auxiliary LLM and dropped from ~90s to ~20ms. Promptware defense scans memory at load time and tags tool results with delimiters to block Brainworm-style indirect injections.

Why does it matter?

Hermes is one of the most-starred agent frameworks on GitHub (171k stars) and Nous routinely rolls research into it before broader release. The velocity rewrite cuts per-conversation function calls by 47% (399k → 213k for a 31-turn chat), which lowers both latency and provider bills. Skill bundles let one slash command load whole workflows, and the Nous-approved MCP catalog gives self-hosters a vetted tool index.

Who is it for?

Self-hosters and indie agent builders running multi-platform Hermes deployments.

Try it

git clone https://github.com/NousResearch/hermes-agent && cd hermes-agent && uv sync

Sources · 2 outlets

Tags

  • agent
  • agent-framework
  • multi-agent
  • kanban
  • promptware-defense
  • open-source
  • mit-license
  • python
  • nous-research
  • self-improving
  • skill-bundles

← All releases · Learn AI