Hermes Agent
The self-improving agent from Nous Research.
An autonomous AI agent that lives on your server, remembers what it learns, and gets more capable the longer it runs. Community guides, skills, and integrations in one place.
Quick Start
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash Installs the Hermes Agent CLI and its Python environment.
Capabilities
Runs on Your Server
Linux, macOS, or WSL2. Runs on a $5 VPS, a GPU cluster, or serverless on Daytona and Modal. Your data and skills stay on your own infrastructure.
Every Chat App
Chat via Telegram, Discord, Slack, WhatsApp, Signal, Email, or the CLI. Start a conversation on one platform and pick it up on another.
Persistent Memory
Long-term memory plus a dialectic user model via Honcho. The longer Hermes runs, the better it knows your projects, preferences, and past decisions.
40+ Built-in Tools
Web search, browser automation, vision, image generation, voice, and multi-model reasoning — all callable in a single conversation.
Real Sandboxing
Seven terminal backends: local, Docker, SSH, Singularity, Modal, Daytona, and Vercel Sandbox. Container hardening, namespace isolation, and command approval workflows.
Self-Writing Skills
Hermes creates skills from experience, improves them as it uses them, and syncs with the open agentskills.io standard. Share what works with the community.
Scheduled Automations
Built-in cron scheduler — describe a job in natural language and let Hermes run it unattended. Daily reports, nightly backups, weekly audits; with `deliver=all` the output fans out to every connected chat platform at once.
Parallel Subagents
Spawn isolated subagents for parallel workstreams. Each runs in its own sandbox and reports back. Write Python scripts that call tools via RPC to collapse multi-step pipelines into one zero-context-cost turn.
OpenAI-Compatible Proxy
Run `hermes proxy` and your Claude Pro, ChatGPT Pro, or SuperGrok OAuth session is exposed as a local OpenAI-compatible HTTP endpoint. Codex CLI, Aider, Cline, and Continue hit it without an API key — one subscription, every tool. New in v0.14.0.
Autonomous /goal Loops
Hand Hermes a target and success criteria with `/goal`. A worker proposes, a separate judge LLM grades — the loop runs until the judge confirms you're done. `/subgoal` adds new criteria mid-flight without restarting. Best fit for tasks with concrete, gradable outcomes.
Live Session /handoff
Mid-conversation, swap to a different model, personality, or profile and `/handoff` migrates the entire session — every message, every tool call, every byte of context — without dropping a thread. New in v0.14.0.
Two entry points, one set of slash commands
Hermes ships a terminal UI you can spin up locally, and a gateway you can chat to from Telegram, Discord, Slack, WhatsApp, Signal, LINE, SimpleX, Microsoft Teams, and the rest of the 22 supported platforms. Most slash commands work the same in both — these are the ones you reach for daily.
| Action | terminal Terminal CLI | chat_bubble Messaging gateway |
|---|---|---|
| Start chatting | hermes | hermes gateway setup, then send the bot a message |
| Start a fresh conversation | /new or /reset | /new or /reset |
| Change the model | /model [provider:model] | /model [provider:model] |
| Set a personality | /personality [name] | /personality [name] |
| Retry or undo the last turn | /retry, /undo | /retry, /undo |
| Compress context / check usage | /compress, /usage, /insights [--days N] | /compress, /usage, /insights [days] |
| Browse skills | /skills or /<skill-name> | /<skill-name> |
| Hand off mid-session | /handoff | /handoff |
| Interrupt the agent | Ctrl+C or send a new message | /stop or send a new message |
| Platform-specific status | /platforms | /status, /sethome |
hermes hermes gateway setup, then send the bot a message /new or /reset /new or /reset /model [provider:model] /model [provider:model] /personality [name] /personality [name] /retry, /undo /retry, /undo /compress, /usage, /insights [--days N] /compress, /usage, /insights [days] /skills or /<skill-name> /<skill-name> /handoff /handoff Ctrl+C or send a new message /stop or send a new message /platforms /status, /sethome Source: NousResearch/hermes-agent README — CLI vs Messaging quick reference.
Hermes Agent In Action
Demos and walkthroughs from the community. See what Hermes can do before you install.
play_circle More videos & community showcase arrow_forwardLatest Updates
From OpenClaw to Hermes Agent: a step-by-step migration
If you've been running OpenClaw, Hermes Agent is its successor. v0.14.0 ships a built-in `hermes claw migrate` that imports your SOUL.md, memories, skills, command allowlist, messaging configs, API keys, and TTS assets. This is the practical walkthrough — what gets imported, what flags you have, and the dry-run-first workflow that avoids surprises.
One Claude Pro subscription, every tool: using `hermes proxy` with Aider, Cline, and Codex CLI
v0.14.0 shipped a one-line command that quietly absorbed a whole class of integration questions: `hermes proxy`. Run it once and you get a local `http://localhost:port` endpoint that speaks the OpenAI API, backed by whichever OAuth provider you're signed into — Claude Pro, ChatGPT Pro, SuperGrok. Aider, Cline, Codex CLI, and Continue all expect an OpenAI-compatible endpoint. They just work with your existing subscription, no API key required. Here's the exact workflow, step by step.
From 7 to 22 in 14 months: every chat platform Hermes Agent runs on
Hermes Agent launched in March 2026 speaking seven chat platforms from a single gateway. Fourteen months later it speaks 22 — from Telegram and Discord to Microsoft Teams, LINE, and SimpleX Chat. Same memory, same skills, same agent across all of them. This is the complete map: every platform by category, the release each one landed in, and how the gateway architecture makes 'one agent, twenty-two front doors' possible.
What People Say
"One config replaced three of our chat bots. This is the one."
@opslead_marc"Hermes on a $5 Hetzner box with load average 0.03. It is absurdly efficient."
@low_latency_dan"The first self-hosted AI assistant I did not uninstall after a week."
@generalist_greg"The ACP integration into VS Code changed how I code. I do not leave the editor anymore."
@yuki_codes"Honcho memory turned a chatbot into something that remembers how I actually work."
@memory_mori"Running Hermes on my old Pixel in Termux. Free, private, always on."
@pocket_engineer"One config replaced three of our chat bots. This is the one."
@opslead_marc"Hermes on a $5 Hetzner box with load average 0.03. It is absurdly efficient."
@low_latency_dan"The first self-hosted AI assistant I did not uninstall after a week."
@generalist_greg"The ACP integration into VS Code changed how I code. I do not leave the editor anymore."
@yuki_codes"Honcho memory turned a chatbot into something that remembers how I actually work."
@memory_mori"Running Hermes on my old Pixel in Termux. Free, private, always on."
@pocket_engineer"Credential pools saved my weekend. Three Anthropic keys rotating on autopilot."
@infra_priya"MCP OAuth 2.1 PKCE plus OSV scanning. Finally an agent framework that takes plugins seriously."
@redteam_blue"Filesystem checkpoints plus /rollback is how I convince skeptics to try agents."
@secdev_sam"My non-technical partner uses Hermes more than I do now. Best onboarding of any AI tool."
@partnered_pat"Self-optimized GPT guidance is the quiet revolution of 2026. They are benchmarking prompts instead of writing them."
@dr_elena_ml"Full MEDIA delivery on Signal means I finally have a private, encrypted line to my assistant."
@private_phoebe"Credential pools saved my weekend. Three Anthropic keys rotating on autopilot."
@infra_priya"MCP OAuth 2.1 PKCE plus OSV scanning. Finally an agent framework that takes plugins seriously."
@redteam_blue"Filesystem checkpoints plus /rollback is how I convince skeptics to try agents."
@secdev_sam"My non-technical partner uses Hermes more than I do now. Best onboarding of any AI tool."
@partnered_pat"Self-optimized GPT guidance is the quiet revolution of 2026. They are benchmarking prompts instead of writing them."
@dr_elena_ml"Full MEDIA delivery on Signal means I finally have a private, encrypted line to my assistant."
@private_phoebeFrequently Asked Questions
Common questions about Hermes Agent, sourced from the official README, release notes, and docs.