Repository navigation
Using Eva

Eva works best when a request states the outcome, relevant context, and any constraints. It can talk through an answer inline, create a downloadable file, retrieve current information, or ask before it acts in the browser, desktop, or another external system.
Use a prompt that makes success testable:
Compare these two approaches for a local-first notes application.
Optimize for offline access, explain the trade-offs, and end with a recommendation.
For current information, name what you need:
Find the weather forecast for tomorrow morning and suggest what to pack.
For a file, ask for a format explicitly:
Create a PDF project brief from this outline.
Eva normally answers in chat. It creates an artifact only when you ask for a file or downloadable format; completed artifacts appear in Assets.
| Area | Use it for |
|---|---|
| Chat / New Chat | Start a clean conversation or continue the current task |
| Sessions | Reopen, rename, pin, or delete saved chat sessions |
| Settings | Models, keys, retrieval mode, prompts, voice, goals, background tasks, cron, and MCP |
| Voice View | Spoken interaction and supported voice controls |
| Agents | Watch subagents, browser/desktop runs, and background activity |
| Assets | Find generated files and changed files from retained coding workspaces |
| Skills | Import, review, edit, enable, disable, or delete reusable instructions |
| Workspaces | Create isolated coding runs and inspect their results |
Eva saves conversations after responses. A new application launch intentionally opens a fresh chat, while existing sessions stay in the Session Explorer. This separates a clean daily start from durable session history.
Provider requests use a bounded, relevant view of the conversation rather than unbounded chat history. System instructions and recent task state are retained where possible, and telemetry records sizes and timings rather than prompt text.
In Settings > General, select the available speech engine and voice, then enable auto-speak only when spoken replies are useful.
- Browser speech: uses the operating system's speech services.
- OpenAI TTS: requires an OpenAI key.
- Amazon Polly: requires AWS credentials.
-
Local Voices: optional local synthesis and transcription installed with
./install.sh --voice-deps; microphone audio stays on Eva's local loopback path instead of being sent to a cloud transcription service.
Use your operating system's microphone permission controls. Eva's optional camera is off by default; see Abilities and Automation.
Ask Eva to find an image when you want relevant public imagery, or to
generate an image when you want a new illustration. Eva detects image
placeholders in responses and either searches Wikimedia Commons or uses
gpt-image-1 for generation, depending on your request and configuration.
Generated artifacts can include PDF, text, CSV, and Markdown. Ask for the format, inspect the result in Assets, and use the provided download/open action. Do not assume a file has been created until it appears as a real asset.
- Goals help track durable intentions. Keep each goal concrete and revise it when the priority changes.
- Prompts offers personality presets and an editable system prompt. Treat a custom prompt as policy, not decoration: keep it short and avoid pasting untrusted instructions from websites or documents.
- Adaptive Review can invoke an independent reviewer for consequential, factual, tool, or complex turns. Routine conversation stays on the selected Eva backend.
Eva may obtain current information or use tools, but it should ask for an in-chat decision before higher-risk actions such as writes, edits, deletes, remote mutations, opaque execution, payments, or sensitive navigation.
Before approving, check the destination, file, or account; make sure the action is no broader than your request; and verify any message, purchase, deletion, or remote change says exactly what you expect.
Read/search/fetch consent can be standing and revocable. Edit, execute, move, delete, and unknown operations require a fresh decision.
Continue with Abilities and Automation when you are ready for Eva to work beyond a single chat response.
Eva AI Assistant is MIT licensed. This wiki documents the current standalone application and its optional integrations. Configuration values, credentials, workspace content, and generated artifacts remain local unless you choose to send them to a configured provider or service.
Start
Configure
Use Eva
Help and Build