AI harness software is what turns a language model into a working agent. Amp is one such harness: an agentic coding environment that connects a language model with tools, cloud sandboxes, and persistent conversations (’threads’), available in the terminal, on the web, and as macOS and iOS apps. Amp is operated by Amp Frontier Corporation, which spun out of Sourcegraph in late 2025. As of 20 September 2026.
What is Amp?
Amp describes itself as a ‘coding agent and development environment built for the frontier’. Originally started as a Sourcegraph project and made publicly available in May 2025, Amp was spun out as an independent company in December 2025. Amp runs across four surfaces: the CLI for macOS, Linux, and Windows (via WSL), the web app at ampcode.com, and native apps for macOS and iOS/iPadOS. The CLI is installed with curl -fsSL https://ampcode.com/install.sh | bash; afterwards, amp starts an interactive session and amp -x "prompt" runs a single, non-interactive pass. The ide connect command from an editor’s command palette additionally links the CLI to VS Code, Cursor, Windsurf, Zed, or Neovim, so Amp sees the open file and selection and edits files with undo support directly in the editor.
Amp as an AI harness
Amp’s harness design shows up in several places. The agent loop evaluates tool requests and runs them in a controlled environment – locally, on a connected runner, or in a cloud sandbox Amp creates called an Orb. Project context comes from AGENTS.md files, which Amp reads automatically in the working directory, parent directories, and system-wide locations; if no AGENTS.md exists, Amp falls back to AGENT.md or CLAUDE.md. On permissions, the documentation is explicit: Amp does not ask for approval before running tools by default – changing that requires writing a custom policy plugin. Amp is extensible via MCP servers (amp mcp add), skills, and plugins, and via specialist subagents: Search, Oracle (deeper analysis as a ‘second opinion’), Librarian (cross-repository research), and Read Thread each work in their own context window and return only a summary to the main agent.
Models and model access
Amp uses several models by default and picks among them automatically depending on the chosen mode (the Dial’s Low, Medium, High, and Ultra levels) – including current models from the Claude family and the GPT family. Model Routing lets you prioritise your own API keys, a ChatGPT subscription, or Custom URL connections to OpenAI-/Anthropic-compatible endpoints and AI gateways such as OpenRouter or Vercel AI Gateway – according to the vendor, currently experimental and in early access, initially for higher-tier and Enterprise customers only. Whether this Custom URL connection is practical for routing through the AI Gateway depends on the specific endpoint requirements and should be checked case by case.
Enterprise use: data protection and governance
Amp’s Minimal Data Retention option limits, according to the vendor, how long connected model providers keep prompts and responses to the time needed to return a response, with exceptions for abuse prevention and legal requirements. Enterprise workspaces get single sign-on via WorkOS, SCIM directory sync, audit logging, and an MCP registry allowlist. Amp names only US-based infrastructure providers; a separate EU processing location is not documented. Important for enterprises: source code and prompts go to whichever model provider is selected and should be reflected in the records of processing activities beforehand.
Orchestrating Amp with herdr
herdr lists Amp in its agent table with the integration level ‘screen manifest’ and no dedicated state or session integration. In practice, this means herdr can detect and display Amp sessions in the terminal but does not fold them into a shared session-management layer across multiple agents.
Where it fits: developer harness and enterprise harness
Amp is a tool for developers working in the terminal, in an IDE-adjacent workflow, or on the go. For non-developer staff, CompanyGPT is the matching harness: the same underlying idea of tools, context, and permissions, but built for enterprise roles, integrations, and audit requirements. The AI Gateway can serve as a shared cost and governance layer across different harnesses. For an overview of other coding agents and orchestrators, see our overview of all AI harnesses. We help companies introduce coding harnesses like Amp safely and connect them to their existing AI governance.