All notable changes to Pier are documented here. This project follows Semantic Versioning.
/connectionswitches between your Pier subscription and your own key in one command, remembering the last model you used on each side instead of making you retype a full model slug.- @-mentions are now honest about what they pulled in. Submit a message with
@README.mdor@src/and a gutter under it shows exactly what was read, listed, or attached, files, directories, pasted images, and skills. - New ultracode mode: turn it on and Pier fans a substantial task out across multiple agents working in parallel instead of working through it sequentially. Toggle with
/ultracodeor theultracodekeyword in a prompt. - Early groundwork for running Pier as a host: a server mode with a sandboxed runner for isolated execution, laying the foundation for running sessions outside your local machine.
- Rewind a session, chat and files together. Press
Esctwice to jump back to any earlier message, edit it, and resend. Pier restores the conversation to that point and can roll the working tree back to how it looked then, reverting edits and bringing back deleted files without touching work you have done since. - Type ahead while the agent is working. Queue up your next messages below the composer, then reorder, edit, or remove them (
ctrl+qto focus the queue). Promote one withEnterto steer the running turn immediately, and it lands cleanly between steps, never mid tool call. - The resume picker is now searchable, with a short transcript preview of each past session so you can find and pick up the right one at a glance.
- A redesigned TUI with a cleaner layout, refreshed permission dialog, and model picker.
- Commands can now reach the network safely. Pier detects when a command needs it (
git push,curl, package installs and more), asks once, and runs it. When something is blocked it steers you to a working path (an HTTPS remote, or an explicit escape hatch for SSH git) instead of dead-ending on a silent failure. - New Monitor watches: point Pier at a shell command, log stream, WebSocket, or HTTP endpoint and have it notify you when something you care about happens, so long-running work can run in the background and ping you when it is done.
- Background work now surfaces cleanly: a completion bell and notice, a live elapsed timer, and a named chip in the footer so you always know what is still running.
- The model picker now shows more at a glance: a
(BYOK)tag on your own-key models, plus vision support and per-million pricing badges (from OpenRouter) so you can compare before you switch. - Cycle permission modes on the fly with
Shift+Tab, mid-turn, without restarting the session. - Richer automation hooks.
PreToolUsecan auto-approve, force a prompt, or rewrite a tool's input before it runs;Notificationfires before an approval prompt;SubagentStart/SubagentStop,PostToolUseFailure, and aFileChangedwatcher are new; and aStophook can re-enter the loop to have the agent re-check its own work before it finishes. Malformedsettings.jsonis now reported at startup instead of being silently ignored.
- First stable release of Pier. The CLI is rebuilt from the ground up as a single, fast binary for macOS, Linux, and Windows.
- Every model is available on every plan, each billed at its provider's API cost: the Pier Hybrid router, Sarvam, DeepSeek, GLM, Kimi, Qwen, Codestral, and more.
- Bring your own Sarvam AI or OpenRouter key and use Pier free, or subscribe to bundle every model under one login.
- The
Bashtool can now request network access on a per-command basis, so commands that need to reach the network are handled cleanly instead of failing silently. - Sharper reliability across longer sessions, with better recovery when a command needs the network and clearer, width-fitting output.
This release is deprecated and no longer supported. Please update to 1.0.0.
- Plan mode now saves your plan to a file as the agent writes it, so you can review or edit the full plan yourself before approving it, not just read it back in the chat.
- When you approve a plan, you can jump straight into Auto mode so the agent starts executing immediately without a second confirmation step.
- Pier is now available on Windows as a native install, alongside macOS and Linux. Install it from PowerShell with
irm https://dl.piercode.com/install.ps1 | iex. - Max plans can now pick a frontier model from a dedicated group in the model picker, so you can reach for OpenAI, Anthropic, Kimi and more on the hardest tasks and fall back to Pier Hybrid for everyday work.
- Plan mode now ends cleanly: the agent presents its finished plan for your approval before it starts making changes, so you always see the full plan first.
- Interactive questions can now accept more than one answer, so the agent can ask you to pick several options at once.
- You can now list and stop the background terminals the agent has open, giving you more control over long-running commands.
- Introduced remote execution: run the agent in the cloud on your repository. It clones your repo, works on it, and pushes its changes back to a branch you can pull, while the CLI stays a thin client you attach from.
- The plan checklist now ticks off each step as the agent finishes it, so you can follow progress in real time instead of seeing everything complete at the end.
- The live token count stays visible while the agent is working, including while it runs commands and edits files.
- Added
/feedbackto send a report (with optional logs) straight to the Pier maintainers from inside a session. - Moved to the new
piercode.comdomains for the API, website, and downloads. - Added a short
sccommand alias, so you can invoke the CLI as eitherpierorsc. - Stability and reliability improvements, including a more dependable self-update.
- First public release of Pier.
- Agentic multi-file edits with a built-in verification loop.
- Gateway planning mode: review the full edit plan before any file is touched.
- Indian-language prompting in English, Hindi, Tamil, Kannada and more.
- Rolling out to early-access developers in small batches.