What happened
ThinkRail’s November 1, 2026 release, described in a blog post by Rinat S, is version 0.3.1. The post highlights a search palette intended to make it easier to recover work spread across projects. Users can open it by double-tapping Shift, pressing Command-K or Control-K, or clicking the search button in the top bar. The empty palette shows waiting work across every open project, ordered by urgency. It can search chats, workspaces, and projects, run commands, and jump directly to a Settings section—for example, typing >prov to go to Providers. ThinkRail says the palette learns which result a user picks. The release post calls this step one toward making the palette the way users do everything in ThinkRail; files and specs are planned next, not included as search targets in this release.
The new provider setup lets users add their own model server under Settings → Providers → Add server. They can choose a preset for Ollama, LM Studio, vLLM, or llama.cpp, or paste an address, then use Find models and save. The configuration is written to pi’s models.json, which means the pi CLI can use it too. ThinkRail says the key never reaches the browser. There are limits: the feature supports the OpenAI chat-completions API only, and the UI does not offer per-model tuning.
Version 0.3.1 also adds package management under Settings → Extensions → Install package. A user can provide a pi install line, a pi.dev or npm link, or a package name. ThinkRail says the interface shows what a package adds before the user enables it. Packages can be switched off, updated, or removed from the same row. The post specifies that disabling a package affects ThinkRail chats only; the pi CLI continues to load it.
The release adds font controls under Settings → Appearance → Fonts for interface and code fonts, including weight, size, and ligatures. Users can choose installed fonts or provide their own files; changes apply throughout the app. Code ligatures are now off by default. Russian joins English, Spanish, Simplified Chinese, and Japanese. The post also lists fixes: a macOS “Maximum call stack size exceeded” crash mid-reply, desktop links opening twice or not at all, a false unread indicator after a background job, and completed subagents incorrectly shown as Working. ThinkRail is available as desktop downloads for macOS, Windows, and Linux; the post also describes installing a local launcher to open a browser UI.
What the archive adds
This is a new fact on a known story, not a first introduction: ThinkRail v0.3.0 added interface translation, GitHub review actions and a merged Files panel. Version 0.3.1 extends that developer-workspace trajectory with search and configuration features; the earlier note does not establish those capabilities.
Trend signal
This release weakly supports harness architecture outlives model choice: ThinkRail exposes a route to configure several compatible model-server options inside the same workspace. That is evidence of model choice as a configurable part of the harness, not proof that production teams can freely swap providers or that the feature works beyond the stated OpenAI-compatible API.
Why now
The release post frames the search palette as the first step toward a single way to do everything in ThinkRail, with files and specs named as future additions. That makes the timing part of a staged product direction, although the post does not give a schedule for those additions.
Why it matters
A workspace that spans several projects can make the active task hard to find. Search across open projects, chats, and workspaces gives developers a direct way to return to work, while command execution and Settings navigation place common actions in the same interface. The model-server setup could also help a team that already runs an OpenAI Chat Completions-compatible endpoint put it into the workspace without exposing its key to the browser, according to the release post.
The package controls address a different operational need: teams can inspect what a pi package adds before using it and manage it through Settings instead of the terminal. The distinction between ThinkRail chats and the pi CLI matters, since turning a package off in ThinkRail does not stop the CLI from loading it. These capabilities improve configuration and navigation; they do not, on the supplied evidence, show better model performance, lower cost, or stronger package security guarantees.
Limits and uncertainty
The source is a product release post, not an independent evaluation. It gives no usage, reliability, latency, or cost measurements for search or model connections, and it does not compare ThinkRail with other workspaces. Provider support is specifically limited to the OpenAI chat-completions API, with no per-model tuning in the UI. The post does not specify how package previews are produced or what security review they represent. It says keys do not reach the browser but does not describe storage or rotation beyond writing server configuration to pi’s models.json. Future search for files and specs is a stated plan, without a date.
Reader takeaway
ThinkRail 0.3.1 is a workflow and configuration update: use the search palette to recover work across projects, and connect a supported model server or manage pi packages from Settings. Check the API compatibility and package behavior carefully—especially the fact that disabling a package in ThinkRail does not disable it in the pi CLI.