Amp is now available as an agent in Linear, allowing users to assign issues directly to Amp to start a new Agent Session. Users can mention @Amp to kick off sessions, and Amp's responses are sent back to the same session with replies along the way. The Agent Session is viewable both in Linear and in the Amp sidebar.

Extended BYOK provider options (OpenRouter, Amazon Bedrock, Google Cloud Agent Platform, Azure Foundry, Vercel AI Gateway, Cloudflare AI Gateway, Ollama Cloud, OpenCode Go, and custom endpoints) to all Amp users on every tier. Previously only available as early access to Megawatt and Gigawatt members. Users can configure these in Model Routing in personal or workspace settings.

Amp now supports Plaid speed for modes that use GPT-6 Astra, leveraging OpenAI's ultrafast speed tier for up to 6× faster inference at 6× cost per token. Users can enable Plaid by moving the speed switch past Fast in the mode picker for supported modes like the default High mode. Plaid is currently limited to Amp-provided inference, with subagents and non-Plaid-enabled models falling back to fast or standard speed.

Claude Opus 5.5 now powers Amp's medium mode by default, replacing GPT-5.6 Sol. Opus 5.5 solved 65% of tasks in internal evals (up from 61% for GPT-5.6 Sol), at 10% lower cost than GPT-5.6 Sol and 25% less than Opus 5. Opus 5.5 also features improved steerability, responding better to mid-task corrections without dropping ongoing work.

Amp now collapses the agent's step-by-step work by default, surfacing results rather than process details. Users can still expand steps when needed, reflecting a shift toward running multiple longer-horizon agents simultaneously rather than closely watching each step.

The Amp app for macOS now automatically starts a runner, eliminating the need to manually run 'amp --no-tui' in a terminal. Users can add folders or projects via App Settings and their Mac will appear as 'This Mac' in the thread picker, accessible from ampcode.com, phone, or Puck. A new 'Keep This Mac Awake' feature prevents the Mac from sleeping while the runner is active and the device is plugged in.

Amp now supports sharing a runner with your entire workspace using the --share flag, allowing all workspace members to spawn agents on a shared machine (e.g., a GPU box, Mac builder, or network-specific dev box). Shared runners receive workspace and project Secrets & Env Vars via --amp-env, but never personal ones. Workspace admins can disable runner sharing via Member Settings.

Amp runners can now create Git worktrees directly from the directory picker, allowing isolated checkouts without touching the main branch. Runners also gained support for Secrets & Env Vars configured on ampcode.com via the --amp-env flag, injecting variables into each thread's shell commands, MCP servers, and plugins. A new 'Archive and Remove Worktree' action cleans up worktree threads when done.

Amp runners can now serve multiple directories simultaneously, either by specifying paths with --dir or automatically discovering Git repositories with --discover-dirs. Directories can be added or removed dynamically while the runner is running without restarting. Runners also now auto-update themselves, checking for new releases approximately once per hour and restarting into the new version at most once every 12 hours.

Amp introduced a free Hobby tier and eliminated monthly plan requirements for users bringing their own compute and model subscriptions/keys. Teams can now invite all members for free with BYOK or ChatGPT subscriptions, with no monthly fees, and power users can upgrade for orb discounts. Zero and minimal data retention policy formally extended to all users and workspaces.

Amp added a Restack button that reorganizes all thread changes into a cleaner set of commits. Users can split large diffs into logical pieces, clean up intermediate commits before merging, or combine small related commits. The final file contents remain unchanged.

Users can now choose the models behind Amp's built-in modes and place custom agents on the Dial. Two new tabs in Settings → Mode Dial allow tuning model and reasoning effort for built-in modes (Tune Modes) and combining built-in modes with custom plugin agents (Build Your Own Dial). Workspace admins can also set a shared dial, with personal choices taking precedence.

Messages sent while the agent is working are now delivered at the next possible opportunity instead of being queued until the agent finishes its turn. This allows the agent to act on user feedback sooner and avoid unnecessary verification steps. Built-in actions like Ship and Review still queue as before, and users can still opt into queuing behavior via Amp or the Amp CLI.

Amp orbs now have an interactive Linux desktop accessible via the Desktop tab in any orb thread. The high-resolution desktop supports use cases like verifying agent work for Linux/Android/Chrome extension development, working with desktop file formats (e.g., .docx in LibreOffice), computer use, CAD, and cloud desktop access. A Fun Games mode is also available for the desktop.