Bitterbot AI

Free

Bitterbot AI is a desktop AI agent that runs entirely on your own hardware. It stores your conversations and learned behaviors in local files you control—not in someone else's cloud.

The system is built around persistent memory. Your agent remembers projects, preferences, and patterns across sessions. It also simulates "dream cycles" that reorganize what it learns, supposedly shaping how it responds over time.

Bitterbot AI sits in the AI Agents category. It is open source, self-hosted, and aimed at technical users who prioritize data ownership over convenience.

Open it if you want a local AI agent with inspectable memory and full control. Skip it if you want something that works in five minutes.

This tool is for people who will trade setup time for ownership.

• Technically skilled users who prioritize data ownership: Your data lives in SQLite and Markdown files you can read, edit, or delete. No vendor holds your memories hostage.
• Developers wanting inspectable agent memory and skill evolution: You can dig into how the agent weights facts, traces identity, and validates skills. The system is built to be opened and understood.
• Households or teams comfortable running their own infrastructure: You will maintain the setup, updates, and troubleshooting. The payoff is no per-seat fees and no cloud surprises.
• Privacy-conscious users willing to trade convenience for control: You skip the slick onboarding and managed hosting. You get agents that never phone home with your data.
• Developers who want persistent context across AI conversations: The agent carries project memory forward. You stop repeating context every time you start a session.
• Self-hosting enthusiasts comfortable with Node and Docker: The installers bundle dependencies, but you still need terminal fluency and patience for rough edges.
• Researchers interested in agent architectures and memory systems: The "hormonal" state model and dream-cycle consolidation are genuinely unusual. Good for studying how agents might simulate continuous identity.
• Power users and system administrators comfortable with CLI workflows: Much of the tool is terminal-native. The skills system rewards those already living in the command line.

These are the parts that stick after you get it running.

• Memory you can actually see: Conversations and learned facts sit in files on your drive. You can back them up, version them, or delete them completely.
• Model choice without lock-in: Plug in your own API keys for Claude, GPT, or run local models via Ollama. Switch anytime without rebuilding your setup.
• Cross-platform installers that handle the pain: macOS and Linux get one-liner scripts with bundled Node. Windows gets automatic WSL2 setup. You still do the work, but the scaffolding is there.
• Dream Engine memory consolidation: Between sessions, the agent replays and reorganizes what it learned. Promoted facts rank higher. Noise fades. It is a real attempt at continuous learning, not just chat persistence.
• MIT-licensed source you can audit: No black box. You can read exactly how the agent weights memories, handles skills, and manages its simulated emotional state.

The setup rewards patience and punishes rushing.

1. Run the one-liner installer for your platform, or pull the Docker Compose for servers.
2. Configure your API keys and model preferences in the Control UI at localhost:19001.
3. Start chatting. Watch memory accumulate in local files you can open and read.
4. Let the agent run its "dream cycles" between sessions to consolidate what it learned.

Skip it if any of these would block you.

• No managed option: You are the sysadmin. There is no hosted version to fall back on when your setup breaks.
• Terminal comfort required: The UI exists, but much of the power and debugging happens at the command line. Non-technical users will struggle.
• Still early and rough: GitHub discussions are sparse. Documentation is thin. You will hit edges that require reading source code to resolve.
• Windows depends on WSL2: No native Windows build. If WSL2 is broken or blocked on your machine, this tool is not for you.
• Self-reported claims, unverified: The P2P skills marketplace and economic layer are more concept than working feature. Treat the roadmap with skepticism.