Self-improving · Postgres-native · Yours to run
The agent with a cognitive substrate
Thoth is a self-improving AI agent built on a five-layer memory substrate that consolidates rather than accumulates. It creates skills from experience, sharpens them in use, and builds a deepening model of who you are across every session — instead of forgetting when the context window closes.
The five-layer substrate · L0 → L4
An MIT-licensed fork of Hermes by Nous Research, maintained by 519lab — rebuilt around the cognitive substrate.
Install
Linux / macOS / WSL2
curl -fsSL https://raw.githubusercontent.com/519lab/thoth-agent/main/scripts/install.sh | bash
Windows (native, PowerShell) — early beta, details →
iex (irm https://raw.githubusercontent.com/519lab/thoth-agent/main/scripts/install.ps1)
Android (Termux) — same curl one-liner as Linux; the installer auto-detects Termux.
See the full Installation Guide for what the installer does, the per-user vs root layout, and Windows-specific notes.
What is Thoth Agent?
It's not a coding copilot tethered to an IDE or a chatbot wrapper around a single API. It's an autonomous agent that gets more capable the longer it runs. It lives wherever you put it — a $5 VPS, a GPU cluster, or serverless infrastructure (Daytona, Modal) that costs nearly nothing when idle. Talk to it from Telegram while it works on a cloud VM you never SSH into yourself. It's not tied to your laptop.
Quick Links
| 🚀 Installation | Install in 60 seconds on Linux, macOS, WSL2, or native Windows (early beta) |
| 📖 Quickstart Tutorial | Your first conversation and key features to try |
| 🗺️ Learning Path | Find the right docs for your experience level |
| ⚙️ Configuration | Config file, providers, models, and options |
| 💬 Messaging Gateway | Set up Telegram, Discord, Slack, WhatsApp, Teams, or more |
| 🔧 Tools & Toolsets | 70+ built-in tools and how to configure them |
| 🧠 Memory System | Persistent memory that grows across sessions |
| 📚 Skills System | Procedural memory the agent creates and reuses |
| 🔌 MCP Integration | Connect to MCP servers, filter their tools, and extend Thoth safely |
| 🧭 Use MCP with Thoth | Practical MCP setup patterns, examples, and tutorials |
| 🎙️ Voice Mode | Real-time voice interaction in CLI, Telegram, Discord, and Discord VC |
| 🗣️ Use Voice Mode with Thoth | Hands-on setup and usage patterns for Thoth voice workflows |
| 🎭 Personality & SOUL.md | Define Thoth's default voice with a global SOUL.md |
| 📄 Context Files | Project context files that shape every conversation |
| 🔒 Security | Command approval, authorization, container isolation |
| 💡 Tips & Best Practices | Quick wins to get the most out of Thoth |
| 🏗️ Architecture | How it works under the hood |
| ❓ FAQ & Troubleshooting | Common questions and solutions |
Key Features
Agent-curated memory with periodic nudges, autonomous skill creation, skill self-improvement during use, Postgres full-text cross-session recall with LLM summarization, and Honcho dialectic user modeling.
Perception → entities → associations → patterns → self-model on Postgres + pgvector, maintained by a roster of always-on sub-agents. Memory that consolidates, not accumulates.
Six terminal backends — local, Docker, SSH, Daytona, Singularity, Modal. Daytona and Modal hibernate when idle, costing nearly nothing.
20+ platforms from one gateway — Telegram, Discord, Slack, WhatsApp, Signal, Matrix, Email, SMS, Microsoft Teams, Home Assistant, and more.
Built-in cron that delivers results to any connected platform on a schedule you set.
Spawn isolated subagents for parallel workstreams. Programmatic Tool Calling via execute_code collapses multi-step pipelines into single inference calls.
Compatible with agentskills.io — skills are portable, shareable, and community-contributed through the Skills Hub.
Connect any MCP server, and search, extract, browse, see (vision), generate images, and speak (TTS) across the open web.
Batch processing, trajectory export, and RL training with Atropos — built by model trainers, forked from Hermes by Nous Research (the lab behind the Hermes, Nomos, and Psyche models).
For LLMs and coding agents
Machine-readable entry points to this documentation:
/llms.txt— curated index of every doc page with short descriptions. ~17 KB, safe to load into an LLM context./llms-full.txt— every doc page concatenated into a single markdown file for one-shot ingestion. ~1.8 MB.
Both files also resolve at /docs/llms.txt and /docs/llms-full.txt. Generated fresh on every deploy.