The eVi platform delivers a professional local-first experience by keeping prompts, transcripts, and keys on disk. The unified core supporting CLI, web, and desktop interfaces simplifies development. Support for multiple local backends and MCP integration enhances its utility. It is a reliable choice for developers seeking privacy and control over their AI infrastructure without relying on cloud services.
Questions about eVi
What is eVi?
eVi is a local-first personal AI assistant software product that lets users chat with a large language model running on their own hardware. It supports tools such as file access, code execution, shell commands, web search, and vision. The same core engine is exposed through a terminal interface, a web app, and a native desktop application.
Which local backends does eVi support and how is the software distributed?
The software is designed to work with locally hosted backends like LM Studio, Ollama, and llama.cpp. It includes features for persistent memory, multi-agent task handling, and MCP server integration. The distribution is available as an open-source project on GitHub and via pip.
What are the three ways to run eVi and how does the desktop app handle updates?
Users can run eVi through a terminal using a fast Typer/Rich REPL, a web app built with FastAPI + SSE, or a signed Tauri desktop application for Windows, macOS, and Linux. The desktop app self-updates from the latest release page and is minisign-signed for the updater.
How does a user install eVi and what are the system requirements?
To get started, users can run 'pip install evi-assistant' or 'evi models recommend' to find a model that fits their hardware. The import package and CLI stay as 'evi', and it requires Python 3.13+ with a local backend like LM Studio or Ollama.
What specific tools and permission controls does eVi provide?
The software offers granular per-tool permissions and a sandboxed run_python feature. Real tools include files, code execution, shell, web search, git, SQLite, PDF, OCR, vision, and voice, all stored locally under ~/.evi.
How does eVi handle memory, skills, and MCP server integration?
eVi supports persistent cross-session memory, on-demand Skills (SKILL.md packets), and a hardware-aware model registry. It can also consume other MCP servers' tools and run as an MCP server exposing curated tools to Claude Desktop and Cursor.
What are the capabilities of the multi-agent and federation features?
The multi-agent feature allows Ultracode fans to work out across parallel subagents that decompose, solve, adversarially verify, and synthesize tasks that one context cannot hold. Federation also lets users delegate heavy work to another eVi on the LAN.
What safety guardrails and local-first principles does eVi implement?
Layered safety includes regex, an LLM judge, and an offline classifier, plus hooks, permission modes, trusted directories, and an MCP allowlist. The system is local-first by default, keeping prompts, transcripts, and keys on the disk.
How does eVi detect available backends and what should users expect regarding code signing?
The website states that the app auto-detects what is live and recommends a model that fits the hardware. It also mentions that SmartScreen/Gatekeeper may warn on first use while OS code-signing is being set up.
What are the core architectural principles and licensing terms of eVi?
The business is described as local-first, with no cloud required. It is built for your own hardware and is MIT licensed. The core engine drives a CLI, a web UI, and a desktop app all over one shared core.
What are the specific package names and version requirements for eVi?
The software is available as a signed Tauri app for Windows, macOS, and Linux. The distribution is evi-assistant, while the import package and CLI stay evi. It requires Python 3.13+.
How does the self-update mechanism work and what security measures are in place?
The website mentions that the app self-updates from the latest release page and that minisign-signed installers are used for the updater. The code signing policy is referenced for details on this process.
What is the official website of eVi?
The official website of eVi is evi-ai.dev.
How is eVi rated on Trustburn?
On Trustburn, eVi has a rating of 3.8 out of 5 based on 5 reviews.
AI-generated from reviews on this page. Verify anything important with the company directly.
It is a functional local AI assistant with good tool integration and solid documentation.
eVi presents a remarkably comprehensive suite of capabilities for a local-first assistant. The ability to run a model on your own machine while using tools for files, code execution, shell access, web search, git, SQLite, PDF processing, OCR, vision, and voice is impressive. The memory and skills system allows for persistent cross-session context and on-demand SKILL.md packets. Federation capabilities let you delegate heavy work to another eVi instance on your LAN, keeping small models for chat while borrowing GPU power. Multi-agent Ultracode fans can decompose tasks across parallel subagents. With support for LM Studio, Ollama, and llama.cpp, it integrates well with existing local setups. The guardrails layer with regex, LLM judge, and offline classifier adds safety. It is a robust, offline-centric platform that truly puts the user in control of their data and hardware.
Having a single Python engine drive the CLI, web UI, and desktop app across platforms offers genuine convenience for a developer workflow. It is nice to chat, run tools, and drive the browser from one core. However, the fact that OS code-signing is still being set up means users must tolerate initial security warnings, which feels like an unfinished polish on an otherwise solid local-first architecture.
The local-first approach appeals to me, yet the installation details for desktop apps suggest potential warnings from SmartScreen or Gatekeeper on first run. The documentation does not clearly specify exactly what is included in the basic distribution versus optional features, so verifying the scope before committing seems wise.
News
No news yet. This is where the business shares its own updates and press.
Photos
No photos yet. Photos shared by reviewers will show up here.
Widgets
Get the widget for your site →About eVi
eVi is a local-first personal AI assistant software product. It lets users chat with a large language model running on their own hardware, with support for tools such as file access, code execution, shell commands, web search, and vision. The same core engine is exposed through a terminal interface, a web app, and a native desktop application. It is designed to work with locally hosted backends like LM Studio, Ollama, and llama.cpp, and includes features for persistent memory, multi-agent task handling, and MCP server integration. The software is distributed via pip and is available as an open-source project on GitHub.
- Website
- evi-ai.dev
- Categories
- Computer software