Amp removes fees and limits for BYOK coding-agent use

• Amp has removed its monthly-seat requirement, BYOK token fees, and BYOK usage limits for non-Enterprise users who supply a model subscription, API key, or compute, per sqs's Free Agent announcement.
• Amp can send inference to a custom URL compatible with OpenAI Chat Completions, OpenAI Responses, or Anthropic APIs, although sqs's endpoint reply says Amp itself remains hosted.
• OpenRouter, Bedrock, Google Cloud Agent Platform, Azure Foundry, Vercel and Cloudflare AI Gateways, Ollama Cloud, OpenCode Go, and custom endpoints are in an experimental rollout that sqs's provider rollout and sqs's early-access clarification initially limited to paid members.
• Ollama Cloud models are already usable through the new routing path, according to ollama's routing post.

The official announcement extends Amp's zero- and minimal-data-retention policy to all users and workspaces, a term previously contractually guaranteed only to Enterprise customers. Amp's Orb docs say a sleeping remote agent retains its conversation, files, and services at zero cost. The runner reference documents a --remote-control-terminal switch for web access to a runner's terminals.

Amp has separated access to the agent from its model-credit and remote-compute businesses. Hobby is free and includes product features, private repositories, external subscriptions and keys; Orbs remain pay-as-you-go remote machines, while a local runner costs nothing to operate.

The paid Individual listing starts at $20 per month with 45,000 minutes of Orb time. Megawatt and Gigawatt members retain average Amp-usage discounts of 60% and 65%, respectively, under the new pricing terms. Teams cost no extra per seat, can mix Hobby and paid members, and receive SAML/OIDC SSO if at least one member is on a paid tier.

Earlier sqs's Hobby update said the tier had shipped the night before, while sqs's tier reply said existing users could switch to Hobby. The Saturday pricing post spread faster than expected, sqs's Saturday reply said; sqs's Florida joke imagined a “Not valid in Florida” disclaimer.

The routing surface is the practical expansion: Amp can steer an agent's inference to a linked subscription, key, gateway, or compatible endpoint rather than treating a single ChatGPT subscription as the only input source.

The announcement's provider list names:

• OpenRouter
• Amazon Bedrock
• Google Cloud Agent Platform, formerly Vertex AI
• Azure Foundry
• Vercel AI Gateway
• Cloudflare AI Gateway
• Ollama Cloud
• OpenCode Go
• Custom endpoint URLs

Custom URLs cover the three API families that sqs's endpoint reply named: OpenAI Chat Completions, OpenAI Responses, and Anthropic-compatible endpoints. Ollama's post says its cloud models already route through Amp, while the broader provider set remains experimental for Megawatt and Gigawatt members before a planned general rollout ollama's routing post.

The no-cost execution path is a user-operated runner. The runner reference says amp --no-tui waits for remotely created threads in its current directory; --runner-id assigns a stable hostname-like ID, and --remote-control-terminal exposes its terminal in the web app.

• A new thread can be assigned to a chosen runner, as sqs's runner selection reply confirmed.
• Two threads on one runner share that runner's directory, just as two local CLI windows launched in the same directory would.
• Amp users can build their own orchestration around local worktrees and additional runners, according to sqs's runner explanation.