Docs /Hermes Capabilities

Hermes Capabilities

Everything Hermes can do: tools, memory, scheduler, API and more.

Hermes v0.14.0 — Full Capabilities

Hermes is not a ChatGPT wrapper. It is an autonomous agent with a closed learning loop that becomes more capable the more it is used.

3 key differences:

Closed learning loop — creates skills from experience, improves them, builds a model of the user
Real autonomy — executes complex tasks without supervision and reports results
Multi-platform native — 22 platforms from a single process

Built-in native tools

Terminal and code

Python, JavaScript, Bash execution
7 backends: local, Docker, SSH, Singularity, Modal, Daytona, Vercel Sandbox
Full file manager (read, write, patch, search)
Integrated Git · Ripgrep for code search

Web and browser

Web search: Firecrawl, Tavily, SearXNG, Brave, DuckDuckGo (no API key)
Browser automation: Playwright, direct CDP — 180x faster in v0.14.0
Web vision: screenshot, visual analysis of pages
Computer use: mouse/keyboard control (works with GPT-4o, Gemini — not just Anthropic)
Scraping: any page, structured extraction

Multimedia

Images: FAL.ai (Flux, SDXL and more)
Text-to-speech: OpenAI, xAI Custom Voices with voice cloning
Transcription: voice → text from Telegram/WhatsApp memos
Video: pluggable backends (Higgsfield, RunwayML)
Visual analysis: image → description, OCR, document analysis

Automation

Cron scheduler: tasks in natural language — “every Monday at 9am send me a summary”
Subagents: delegate tasks to parallel agents in their own context
Multi-agent Kanban: board with heartbeat, zombie detection, automatic retries
MCP: connect any MCP server with OAuth 2.1

Memory System

Type	What it remembers
UserMemory	Who you are, your preferences, working style, team
EntityMemory	Companies, clients, projects, third-party systems
LearnedKnowledge	Complex procedures it solved before
DecisionLog	Why it made certain decisions (auditable)
SessionContext	Multi-step work plan within a session
FTS5 Search	Semantic search across all past conversations

Scheduled automations

The native scheduler understands natural language:

"Every Monday at 8am: analyze news in the retail sector and send me a summary"
"Daily at 6pm: if BTC dropped >5% notify me"
"On the 1st of each month: generate sales report from Airtable and send it by email"

Delivery via any platform: Telegram, WhatsApp, Discord, Email.

API Server (port 8642)

When API_SERVER_ENABLED=true, Hermes exposes an OpenAI-compatible API:

GET  /health                    — health check
GET  /v1/models                 — available models
POST /v1/chat/completions       — Chat Completions (stateless)
POST /v1/responses              — Responses API (stateful)
POST /v1/runs                   — run async
GET  /v1/runs/{id}/events       — SSE stream
POST /v1/runs/{id}/approval     — resolve approval

Any OpenAI-compatible client can connect: Open WebUI, LibreChat, Cursor, Aider.

v0.14.0 exclusive capabilities

OpenAI-compatible local proxy — Cursor, Codex or Cline can use Claude Pro or Grok models without additional API keys
Microsoft Teams end-to-end — read and reply to Teams messages
Computer use with non-Anthropic models — GPT-4o and Gemini control the desktop
Cross-session Claude prompt caching — -60% cost on long conversations
xAI Grok 4.3 at 1M tokens of context
LSP semantic diagnostics — verifies semantic errors in code before ending the turn
/handoff — transfers an active session to another model without losing context
pip install hermes-agent — available on PyPI

← Previous

What is martes.app?

22 Messaging Platforms