CueCode

CueCode is a fork of Zed — a spec-driven agent IDE with local/BYOK models by default, no zed.dev sign-in wall, and separate user data from Zed.

Upstream: Zed (GPL-3.0-or-later). See About in the app for full attribution.

---

Specs & build plan

Product and engineering specs live in the workspace .cursor/specs/ tree:

• Spec index
• Fork & rebrand
• Master build plan
• Implementation roadmap

Phase 0 exit criteria and QA scripts are in 03-fork-and-rebrand §manual QA.

---

Developing CueCode

From this directory (CueCode-IDE/):

cargo run -p cuecode --bin cuecode

Or use the local helper:

./script/cuecode-local

Fresh config paths (no collision with Zed):

Platform	Config / data
macOS	~/Library/Application Support/CueCode/
Linux	~/.config/cuecode/
Windows	%LOCALAPPDATA%\CueCode\

Quickstart

# 1. build + run (first build is slow — thousands of crates)
cd CueCode-IDE
cargo run -p cuecode --bin cuecode

# 2. open a project at startup
./target/debug/cuecode /path/to/project

# 3. run the agent panel: Cmd+R (macOS) / Ctrl+R (linux) → "agent: open"

Config lives in assets/settings/default.json (defaults shipped with the app) overlaid by user settings in the per-platform config dir above. Edit user settings via Cmd+, → Settings Editor, or by editing the JSON directly.

For the canonical dev loop with the QA fixture:

# from monorepo root
./script/clone-qa-fixture-repo
cd CueCode-IDE && ./script/run-cuecode-dev ../cuecode-testing-repo

Bring your own key (BYOK)

CueCode is bring-your-own-key: the agent and inline edit predictions run against your own model providers — no zed.dev sign-in wall, no hosted billing.

• Agent chat — configure an OpenAI-compatible provider (GLM via Z.ai, OpenAI, Ollama, etc.) under language_models.openai_compatible in settings. Add the API key via the in-app keychain (Settings → LLM Providers).
• Edit predictions (inline Tab completions) — by default reuse the agent's configured OpenAI-compatible provider, so a single API setup powers both chat and inline predictions. The provider endpoint must support legacy /completions (FIM-style) requests; a coder/FIM model gives the best results.

To use a dedicated endpoint/model for predictions:

{
  "edit_predictions": {
    "provider": "open_ai_compatible_api",
    "open_ai_compatible_api": {
      "api_url": "https://your-provider.example/v1",
      "model": "your-coder-model",
      "prompt_format": "infer",
      "max_output_tokens": 64
    }
  }
}

Full setup, modes (eager / subtle), and provider list: docs/src/ai/edit-prediction.md and docs/src/ai/llm-providers.md.

Agent UX defaults

Shipped defaults (assets/settings/default.json, agent section) are tuned for a Cursor-like compact streaming experience:

• thinking_display: "preview" — compact thinking blocks
• expand_edit_card: false — edit cards collapsed by default
• expand_terminal_card: false — terminal cards collapsed by default

Agent surfaces

• Agent panel — Threads (chat) and Plan surfaces; see docs/src/ai/agent-panel.md.
• Planning Hub / Layout Studio — spec-driven build plans, multi-column PulseBoard layout. Specs: ../.cursor/specs/design/16-planning-hub.md, ../.cursor/specs/design/17-layout-studio.md.
• Terminal threads — docs/src/ai/terminal-threads.md.

---

CLI

cargo build -p cuecode --bin cuecode
./target/debug/cuecode --help    # product name should say CueCode

Install CLI to PATH (macOS, from the app): Command Palette → install cli.

---

Packaging

Platform	Script
macOS	script/bundle-mac
Linux	script/bundle-linux
Windows	script/bundle-windows.ps1
Flatpak	script/flatpak/bundle-flatpak

Rebrand regression gate:

./script/rebrand-check.sh

Phase 0 smoke (automated QA-P0):

./script/qa-p0.sh

Runs with --user-data-dir /tmp/cuecode-qa-* so your real CueCode config is not modified.

---

Contributing

See CONTRIBUTING.md. Upstream Zed contribution docs remain below for reference where the fork has not diverged.

---

Licensing

CueCode is based on Zed and inherits GPL-3.0-or-later obligations. Third-party license compliance uses script/generate-licenses and script/licenses/zed-licenses.toml (upstream naming).