raylee-ai/lancer

Lancer is an Alpha desktop workspace built on pi-gui and pi, aimed at people who run several AI-assisted tasks at once and need to see how each one reached its result. Instead of a single chat log, it places sessions, branches, run activity and delivered files in one view. It is a finished-style application for personal or team working, not a library you embed in other AI applications.

How a workflow looks

A helper can manage multiple conversations at once. A task can spawn child and grandchild branches, and results return to the parent conversation for consolidation. The right-hand GitTree shows several conversation lines, branches, merges and delivery nodes together, and you can focus the current conversation. The README is explicit that this is a task-event graph, and not every node corresponds to a Git commit.

Run feedback is visual: active bubbles and process lines get a flowing border, and model or delivery actions send same-colour rays to the matching node. Models and accounts are connected through pi, covering subscription accounts, API keys and custom endpoints; model brands and gateways are managed separately.

A documented video workflow downloads or receives video, extracts key frames, and produces evidence-backed analysis and handoff files. The author warns that automatic replication is not guaranteed to match a template one-to-one; Doubao takeover, model analysis, code animation and final editing are separate steps that need separate acceptance. Doubao takeover is macOS-only, needs local accessibility permission, and may break when the desktop UI changes.

Using only the ray UI

The light UI package is a separate path with different requirements: it provides single-line, fork-and-merge and multi-session examples and does not depend on Electron, the CLI or a model service. It can be exported to a new directory and run with npm, so this visual piece alone does not require the full desktop stack.

Requirements and limits

Building from source needs Node >=22.19.0 <26 (24 recommended), Git and pnpm 10.34.9; macOS also needs Xcode Command Line Tools, while Linux needs Python 3, make and a C++ compiler. macOS Apple Silicon is the main validated platform, and Windows native behaviour was not validated in this release round. The source archive contains no ready-to-run .app, and no standalone app download has been released yet; do not treat the source package as an installer.

Video processing additionally requires working FFmpeg, FFprobe and yt-dlp. Starting the conversation interface alone does not need those media tools.

For a local macOS build, the script requires a persistent signing certificate, and it will not create one, switch to a temporary signature or notarize. A local signed app is not a notarized public release. The installer keeps a recoverable previous version and does not clear conversation data, but do not replace an installed app until the candidate is verified.

Session and account data use pi's ~/.pi/agent; macOS UI state is under ~/Library/Application Support/Lancer. A separate source folder does not isolate runtime data, so fully isolated manual testing requires a separate OS user. The project is MIT-licensed, with upstream pi-gui v1.0.1 notices retained, but third-party model trademarks, avatars, reference videos and Fate assets are not automatically redistributable under that licence README repository platform validation.

HotRepo's take

The only supplied observation is 101 total stars on the NEW-REPOSITORY board on 2026-10-31, where it ranked #22 in TypeScript. That is a cumulative snapshot, not daily growth, and there is not enough comparable history to judge momentum.