Zed AI Editor On Linux, A Real Setup For Ubuntu 24.04
Zed Industries' Rust-built editor with Claude built in, my install on a stock ThinkCentre
Zed is the editor I keep coming back to when I want responsiveness over feature breadth. Built in Rust, GPU-accelerated, with first-class AI chat and inline-edit, it ships with Claude integration and a clean prompt-completion flow. The official Linux build needed two tweaks on my Ubuntu 24.04 box before it was usable; this is the install that worked.
What you'll build
Zed installed on Ubuntu 24.04 LTS via the official tarball, signed in for AI features against your Anthropic credit, with the Wayland-safe launch flag that the docs do not surface. Roughly 10 minutes.
Caption: Zed editor with the assistant panel open against Claude.
Prerequisites
- Linux x86_64 (Ubuntu 22.04+, Fedora 39+, Debian 12+)
- 4GB RAM minimum, 8GB comfortable
- Vulkan-capable GPU. Most Intel iGPUs from 2018+ work fine
- Anthropic API credit if you want AI features. Zed offers its own subscription tier as an alternative
If your Linux box is X11, the install is simpler. Wayland needs one extra step.
Step 1, install Zed
curl -f https://zed.dev/install.sh | sh
zed --version
The script writes the binary to ~/.local/bin/zed and a desktop file to ~/.local/share/applications/. Logout and login once for the desktop entry to refresh.
Step 2, fix Wayland launch (if applicable)
On Ubuntu 24.04 with Wayland (the default), Zed launched but rendered black until I clicked. The fix is one env var:
echo 'export ZED_DISABLE_WAYLAND_ATTACH=1' >> ~/.bashrc
source ~/.bashrc
If you are on X11 (still common on older Ubuntu and most Fedora installs), skip this step.
Step 3, sign in for AI
Open Zed, hit Cmd+? (or Ctrl+? on Linux), pick "Open Settings". Add your Anthropic API key under language_models.anthropic.api_key or use the GUI sign-in flow. The sign-in flow opens a browser, redirects to zed.dev, attaches your account.
For Claude access, you can either supply your own Anthropic key or subscribe to Zed Pro at $20/mo, which routes through their account.
Step 4, run an inline edit
Open any TypeScript file, select a function, hit Ctrl+Enter. The inline-edit popup appears:
> add a try-catch around the JSON parse and log a warning on failure
Zed shows the proposed change as a diff in-place. Hit Enter to accept, Esc to dismiss.
Step 5, use the assistant panel
Cmd+R (Ctrl+R on Linux) opens the assistant panel. Drop a file with @file, ask a question:
@src/lib/news-loader.ts -- where does loadNews handle parse errors right now?
The assistant reads the file inline and answers without leaving the editor.
First run
A typical 5-minute Zed session for me:
[open repo] zed ~/projects/my-app
[Ctrl+Shift+P] command palette
[type: assistant -> opens panel]
[type: @src/components/NewsCard.tsx -- explain the data flow in this component]
[Claude responds in the panel; I read]
[click into the file, select a block, Ctrl+Enter for inline edit]
The whole thing feels closer to vim than to VS Code. If you want a heavy IDE, this is not it.
What broke for me
Two Linux-specific issues. First, font rendering on a HiDPI external monitor was blurry until I added "ui_font_size": 18 and "buffer_font_size": 16 to ~/.config/zed/settings.json. Zed defaults assume a low-DPI 1080p display; on my 4K external the defaults are tiny and antialiasing looked off until I bumped sizes.
Second, the AppArmor profile on stock Ubuntu 24.04 blocked Zed from spawning the Rust analyzer for some projects. I added a profile override at /etc/apparmor.d/local/zed allowing the binary to spawn child processes, then sudo systemctl reload apparmor. The error in Zed was a vague "language server unavailable"; the AppArmor logs in /var/log/syslog made it obvious.
What it costs
| Item | Cost |
|---|---|
| Zed editor | Free (Apache 2.0) |
| Zed Pro (managed AI) | $20/mo |
| BYO Anthropic API | Pay per use |
| BYO local Ollama | Free |
For an Indian developer wanting Claude in-editor for under Rs 1,700/mo, Zed Pro is the cleanest pricing. If you already have Anthropic credits, BYO key is cheaper for casual use.
When NOT to use this
Skip Zed if your work needs heavy plugin market, language servers for less-common languages (think OCaml, Nim, Roc), or you depend on a debugger UI. Zed is intentionally minimal; it has the languages and features the team prioritises and not much more. VS Code or JetBrains is a better fit there.
Skip if your machine is older than ~2018; Zed's GPU rendering needs Vulkan support that older Intel iGPUs do not always provide. On an Nvidia box, if Zed and video playback micro-stutter, the cause is usually the GPU dropping its clocks, not the editor, and the fix is one line.
Indian operator angle
Zed's perf advantage matters most on lower-spec machines, exactly the boxes Indian dev shops actually run. On a 4-year-old laptop with an Intel iGPU, Zed feels twice as responsive as VS Code with the Continue extension. For a Tier-2 city dev shop with mid-range hardware, that compounds across a workday.
Billing for Zed Pro is in USD; same forex story as the other paid tools. The BYO Anthropic key path lets you treat Zed as a free editor and pay only for the AI minutes you use; that pricing model fits Indian-operator economics better than a flat subscription.
Related
More AI Coding
Building a Custom MCP Server in Python: Claude Reaches My Stack
Claude Code is sharp until it hits the edge of your machine and your private tools. I wrote three small MCP servers in Python to close that gap. Here is the real pattern, the real gotcha that bit me, and what it costs.
Claude Code Subagents in Practice: Fork Flag, Cache Leak, Worktree Trap
Fanning out subagents in Claude Code looks free until you hit the cap or your forks clobber each other's commits. These are the real fixes I learned running fanouts: the fork env flag that shares the parent's cache, the WebFetch cache leak, and the worktree pattern for parallel writers.
I Gave My AI Agents a Memory With SQLite FTS5 (No Vector DB)
Most agent-memory setups reach for Pinecone or pgvector by reflex. I put 2000+ markdown files behind SQLite FTS5 with BM25 ranking, and my agents now answer their own 'who is X' questions in under a second for zero tokens. Here is the schema, the query, and the one place lexical search loses.