Amp Review, Pricing & Alternatives (2026)

Amp is a terminal and web coding agent made by Amp Frontier Corporation, which became independent from Sourcegraph in December 2025. Its main unit is the Orb, a remote computer where an agent receives a repository, tools, and a task. Users can resume threads from the web, terminal, or phone, and agents can continue after the client closes. Amp also supports runners on a user’s own computers. The company publishes product and pricing information, but it does not publish a standalone user count or funding total on the cited pages. The product is aimed at developers who want persistent agent work, remote execution, and access to several model providers rather than a single model subscription.

A typical Amp workflow starts with a repository and a thread. The agent can inspect files, use tools, edit code, run tests, and work through configured GitHub integrations. Each thread can use an Orb, which separates agent work from the developer’s local machine and can keep running when the client is closed. Amp supports shared threads, multiplayer work, permissions, portals, plugins, skills, analytics, pooled credits, and model routing for teams. The terminal client provides configuration and usage commands, while the web interface provides thread history and remote control. Orb size affects usage speed. Amp also supports self-hosted Orbs and user-owned runners, giving teams a way to keep compute outside Amp’s hosted infrastructure.

Amp supplies the agent, tools, thread interface, and Orb execution layer, while inference can come from Amp or a configured provider. Its routing documentation lists Anthropic, OpenAI, Google, Meta, xAI, Bedrock, Vertex, Azure AI Foundry, OpenRouter, and other compatible endpoints. Users can select models and modes through the Dial or configure routing rules. The company’s September 2026 Free Agent update made BYOK and linked subscriptions available without Amp token fees or limits, apart from Enterprise terms. The pricing documentation mentions GPT-6 and GPT-5.6 subscription access, although model availability depends on the connection. This split matters: remote execution, model inference, and provider retention are separate parts of the workflow.

Hobby costs $0 per month and includes all product features, pay-as-you-go Amp-hosted Orbs, unlimited free self-hosted Orbs, linked subscriptions, and BYOK without Amp token fees or limits. Individual costs $20 per month and includes 45,000 minutes of Amp-hosted Orb time, plus unlimited self-hosted Orbs and the same provider options. Teams has no extra workspace charge, supports members on different tiers, and adds shared threads, permissions, billing, pooled credits, analytics, and shared routing. Enterprise is custom and adds SCIM, spend controls, audit and network controls, regional endpoints, higher quotas, and priority support. Model usage can consume paid credits, while Orb minutes are billed separately. Purchased credits expire after twelve months.

Amp’s strongest distinction is persistent remote work, but its billing is more involved than a single editor subscription because model credits and Orb minutes are separate. Users sending code through external providers must also consider those providers’ retention and training settings. Amp says it does not train on User Content without opt-in, and its terms prohibit sensitive regulated data. Claude Code is a closer terminal comparison, Codex CLI is more focused on a local OpenAI workflow, and Cursor centers on an editor. Choose Amp for remote parallel agents, shared threads, self-managed runners, and provider flexibility; choose the alternatives when local execution or an editor-first workflow matters more. Who should pick it: developers and teams needing persistent remote agents.