An AI assistant that sees through your camera, controls your browser and desktop, remembers everything, and learns from experience. Local-first. Open source.
Voice, vision, browsing, desktop control, memory, and autonomous task scheduling. No framework, no build step.
Eva sees through your webcam. Face-detection auto-wake, on-demand "look" commands, ambient presence sensing. Frames stay local unless you ask her to describe what she sees.
Controls a real Chromium browser through the DOM. Persistent login, precise click-by-element, hybrid vision fallback. Shops, fills forms, navigates, pauses for confirmation on purchases.
Drives your actual mouse and keyboard to operate any application. Focus windows, open apps, click buttons. Optional AT-SPI semantic control via computer-use-linux MCP.
Full-screen voice orb with wake/barge-in, echo management, spoken progress narration. TTS via OpenAI, AWS Polly, browser, or Bark. Eva speaks her results, not just displays them.
Kusto/ADX-backed conversation history, emotion tracking, knowledge graph, semantic recall. She remembers what you talked about last week and how you were feeling.
Auto-extracts reusable skills from successful complex tasks. Skills are stored as drafts for review and applied automatically to matching future requests.
Schedule recurring tasks with standard cron expressions. Morning briefings, periodic market checks, nightly reports. Eva runs each prompt through ACP at the scheduled time.
Spawn up to 4 isolated subagent tasks that run concurrently. Results delivered via notifications. Break complex work into parallel streams.
OpenAI, Google Gemini, GitHub Copilot, local models via lm-studio. Switch providers from a dropdown. No code changes, no lock-in.
Structured readiness report for every subsystem: ACP, browser, desktop, camera, Kusto, cron, cognition. Tells you exactly what's working and what needs fixing.
Azure, GitHub, Kusto, and computer-use-linux MCP servers configurable from the UI. Extend Eva's capabilities with any MCP-compatible tool server.
Eva + Reviewer dual-agent pipeline with configurable models and max review cycles. Every response is drafted, critiqued, and refined before you see it.
No build step. No framework. Pure HTML, CSS, and JavaScript in the browser, backed by a Python bridge that connects to any LLM provider.
| Eva | Hermes Agent | ChatGPT Desktop | Claude Desktop | |
|---|---|---|---|---|
| Open source | Yes (MIT) | Yes (MIT) | No | No |
| Camera/vision | Built-in | No | No | No |
| Browser automation | Built-in (DOM + vision) | Via MCP | No | Via MCP |
| Desktop control | Built-in + MCP | Via MCP | No | Via MCP |
| Voice interface | Full voice view | Transcription only | Yes | No |
| Self-improving skills | Auto-learn | Yes | No | No |
| Scheduled tasks | Cron | Cron | No | No |
| Persistent memory | Kusto/ADX | FTS5 + files | Limited | Limited |
| Multi-provider | 4+ providers | 200+ models | OpenAI only | Anthropic only |
| No build step | Yes | pip install | N/A | N/A |
One command. No API keys required to start.