Refact.ai

Refact.ai started as a code-completion tool and has evolved into a full autonomous coding agent that plans multi-step changes, edits files, runs commands, and verifies the result against tests — all from inside your IDE. The differentiator versus closed competitors is that Refact ships its agent runtime as open source under a permissive license, and the platform supports both cloud and fully self-hosted deployments on your own GPUs.

The IDE plugins cover VS Code and JetBrains IDEs, exposing a chat sidebar, inline completions, and an Agent mode that takes a natural-language task ('refactor the auth middleware to support API keys'), proposes a plan, executes it across multiple files, and shows a diff for approval. Refact supports a wide range of models including Claude, GPT-4-class models, DeepSeek, Qwen, Llama, and the company's own fine-tuned models, with automatic routing between fast/cheap models for autocompletion and frontier models for agent steps.

For enterprises, the self-hosted option is the headline. Refact can run inside a VPC or on bare-metal GPU clusters, never sending code to a third-party API, which satisfies financial-services and government compliance requirements that block tools like Copilot. The self-hosted edition includes fine-tuning so the model learns your codebase patterns over time, and supports SSO, audit logs, and role-based access.

Refact added native Model Context Protocol (MCP) support in 2025, meaning developers can plug in MCP servers (filesystem, GitHub, Postgres, Slack, custom) and Refact's agent will use those tools during reasoning. This makes Refact a strong choice for teams that want a code-aware agent but also want it to write to ticketing systems, query production databases, or trigger CI runs as part of a single task. Pricing has a free individual tier and paid team/enterprise plans with self-hosting available on annual contracts.