Donovan Agent

The self-learning computer agent

Built for development sessions where the agent needs context, tools, memory, and time.

Debug a local projectRun checks and summarize failuresCreate reusable agent skillsSchedule recurring workSearch, inspect, and verify

Get started

One line install.

Choose your platform, paste the command, then run donovan to launch setup mode.

macOS / Linux

curl -fsSL https://raw.githubusercontent.com/tudor-22/donovan-agent/main/install.sh | bash

What People Say

Get started

"It actually understands my whole project, edits files properly, runs tests, and remembers what we did yesterday."

Alexandru Herman

"Installed Donovan in 30 seconds with the one-liner and immediately had it refactor a messy module while I grabbed coffee."

Andreea Roman

"It's quite a good computer agent for an early release, excited for the next version."

Alex Plesa

"Using Donovan with Ollama local on a private repo working 24/7 and fixing bugs."

Chloe Chen

"As a solo founder, Donovan has become my entire backend team. Context memory is actually good. Skills learning is actually useful."

Tom Capone

"I like the option to add your own 3rd party models to Donovan Agent. Makes vLLM integrate easily."

Artie Jason

What it does

An agent that can use the machine, not just describe it.

Acts on the computer

Runs commands, edits files, navigates projects, and uses tools from the terminal.

Learns reusable skills

Turns repeated successful work into reusable workflows and project conventions.

Keeps useful memory

Stores patterns, preferences, troubleshooting notes, and working context.

Respects boundaries

Uses approvals, checkpoints, audit logs, and risk checks for sensitive actions.

Highlights

Everything needed for a local computer agent.

Terminal-first interactive agent built with Rich and prompt_toolkit

Provider support for OpenAI, Anthropic, DeepSeek, Qwen, LM Studio, and Ollama

Local file tools for reading, writing, patching, listing, and searching

Shell execution with platform-aware command handling

Local Python execution for scripts and analysis

Tavily web search integration

Browser automation with optional Playwright support

MCP integration for external tools, resources, and prompts

SQLite-backed sessions, messages, tool calls, audit logs, memories, and learned skills

Planning, thinking summaries, scheduled tasks, checkpoints, subagents, and activity stream

Approval gates for risky tool use

Providers

Donovan supports hosted, local, and OpenAI-compatible model endpoints.

OpenAIOllamaAnthropicDeepSeekQwenLM Studio/v1Custom local or hosted /v1/chat/completions endpoints

How it improves

The loop is simple.

Understands the project, task, tools, and previous context.

Takes action through files, shell commands, browsers, MCP tools, and scheduled runs.

Checks results with tests, command output, screenshots, and session history.

Saves what worked so future sessions become faster and more consistent.