Getting Started

Getting Started

This guide gets a working Eva installation from zero to a first useful task. Start with the standalone desktop app unless you are developing Eva itself.

Before You Install

Goal	What you need
Eva with cloud-backed AIG	Linux, Node.js 24+, Python 3.12+, and GitHub Copilot CLI signed in locally
Fully local chat	LM Studio with a local OpenAI-compatible model; no cloud model key required
Direct OpenAI, GitHub Models, or Gemini	An account and key for the selected provider only
Coding Workspaces and ACP tools	GitHub Copilot CLI authenticated with copilot auth login
Browser, desktop, camera, Signal, or Local Voices	The optional prerequisite for that feature; configure it only when needed

Eva's standalone runtime keeps mutable state under ~/.config/eva-standalone by default. Set EVA_CONFIG_DIR before launching if you want a different local state location. The rebuildable installed source is kept separately under ~/.eva on Linux.

Install the Standalone App on Linux

curl -fsSL https://appatalks.github.io/eva-agent/get-eva.sh | bash
eva

The installer adds Eva to your application menu as well. Use the app menu or the eva command for subsequent launches.

First Launch Checklist

1. Open Settings.
2. In Models, select Eva (AIG) for Eva's integrated memory, persona, retrieval, and optional review pipeline.
3. In Auth, configure the one provider route you want to begin with: Copilot ACP, OpenAI, GitHub Models, Gemini, or none for LM Studio.
4. Return to chat and ask for a bounded, observable task, such as: "Summarize this plan and identify the next three decisions."
5. When Eva requests an approval, read what will happen and choose Allow once only when it matches your intent.

Eva starts a fresh chat on each launch. Previous sessions remain available in the Session Explorer, so a clean first screen does not delete your saved work.

Pick Your First Route

Integrated Experience: Eva (AIG)

Use Eva (AIG) when you want Eva's full orchestration: structured memory context, relevant data retrieval, persistent goals and skills, emotion tracking, and optional adaptive review. Its selected backend can use Copilot ACP, OpenAI, GitHub Models, or LM Studio. See Models and Providers.

Local-Only: LM Studio

1. Start an OpenAI-compatible model server in LM Studio.
2. Confirm it is listening on the configured local endpoint, normally http://localhost:1234/v1/chat/completions.
3. Choose an LM Studio model and set data retrieval mode to Local.
4. Send a simple prompt, then add local MCP tools only when you need them.

Eva's local SQLite memory remains available in this route. Details are in Local Mode and MCP.

Copilot ACP

For Copilot-backed chat, tools, retrieval, subagents, and coding workspaces, install and authenticate the CLI on the same computer:

copilot auth login

Then choose a Copilot ACP backend in Settings. Eva speaks to the CLI through a loopback bridge; it does not need you to paste the CLI credential into the UI.

Configure Only What You Use

Setting area	Typical first configuration
General	Theme, speech engine/voice, auto-speak, local/cloud retrieval mode
Models	Eva backend, provider model, reasoning effort, adaptive review
Auth	A provider key, GitHub authentication, or optional Signal details
Prompts	A personality preset or a carefully scoped system prompt
MCP	Optional Kusto, GitHub, Azure, or custom tool configuration

Credentials are local configuration. Do not paste API keys, personal access tokens, or private data into chat, Skills, issue reports, or screenshots.

Verify the Installation

Open Settings > General and confirm the app reports the expected data mode. Then ask Eva for a non-sensitive task. If you selected a provider requiring authentication, a clear provider error is preferable to a silent fallback; use Troubleshooting to diagnose it.

For source builds, AppImage details, and Windows packaging, see Developer Guide.

Next Steps

• Using Eva for chat, sessions, Voice View, and settings habits.
• Memory, Learning, and Privacy before enabling persistent or autonomous workflows.
• Abilities and Automation when you want Eva to act outside chat.