aether · rust · terminal · ai
Aether is a keyboard-first AI agent that lives in your shell. A TUI chat over your code, an MCP server for your tools, and gist or folder sync — in one native Rust binary. No Node runtime. No browser tab. No noise.
curl -fsSL https://raw.githubusercontent.com/Aetherdz/aether/main/scripts/install.sh | bash
git clone https://github.com/Aetherdz/aether && cd aether && cargo build --release && ./target/release/aether tui
all builds + SHA256SUMS.txt on GitHub Releases
measured 2026-08-09 · 5-run median · benchmark/run.sh — reproducible in this repo
quick start
One prebuilt native binary for your OS. No Rust toolchain, no Node, nothing else.
# downloads + verifies SHA-256, installs to ~/.local/bin
$ curl -fsSL https://raw.githubusercontent.com/Aetherdz/aether/main/scripts/install.sh | bash
The free zen provider works out of the box — no API key needed.
# ask a one-shot question
$ aether ask "explain this repo"
Keyboard-first UI: session list, chat, and a live plan→build→route agent screen.
# full terminal UI
$ aether tui
capabilities
tui
A ratatui interface with a chrome header (model, provider, version), a tab bar (chat · agent · sessions), wheel scroll, and a live token ledger. All from the keyboard.
agent
Run the three-model agent loop straight from the TUI. Three live panels show the planner, builder, and router as they work — plan cards, todo progress, and per-step verdicts update in real time, with live counters on every panel.
mcp
Speaks Model Context Protocol over stdio and streamable-HTTP. Your editor, your agents, and Aether share one context layer.
sync
Sessions sync to GitHub gists or a local folder, merged line-by-line with origin fingerprints — no silent overwrites on concurrent edits.
native
No Node runtime, no bundled Electron, no 400 MB install. Compiled, typed, and fast — the way a terminal tool should be.
ledger
Every request is counted. A right-side rail tracks tokens in, tokens out, and total live — you always know what a session cost before you commit to it.
polish
An empty chat opens on the logo screen with a blinking cursor, and your terminal title is restored on exit — no dead space, no leftover state.
faq
No. The free zen provider is built in and works out of the box — hosted by opencode.ai. Bring your own key or a local model anytime.
Honest answer: the free zen endpoint is hosted by opencode.ai and sees your conversation payload. Local models (Ollama, LM Studio) run fully offline. No telemetry, no accounts — see docs/zen-privacy.md.
No. One native Rust binary — no runtime, no interpreter, no 400 MB install. That is the whole point.
Yes — via the agent loop with path-sandboxed tools and a command approval gate: every write shows a diff preview and waits for your y/N before touching disk.
Yes — MIT licensed. Public, auditable, and free. Fork it, vendor it, build on it.
honest comparison
| — | Aether | Node-based agents |
|---|---|---|
| Runtime | One native Rust binary | Node + hundreds of MB of deps |
| Binary size | 4.99 MB — one stripped binary | hundreds of MB installed |
| Startup | ~112 ms cold start | 1 s+ cold start per run |
| Agent loop | plan → build → route, 3 models, sandboxed tools | varies by tool |
| Sessions as files | Plain JSONL — grep it, script it, own it | Often bespoke databases |
| MCP server | Built in — stdio + Streamable HTTP | Requires separate config |
| Telemetry | None — verifiable | Varies |
Aether figures are measured in this repo (benchmark/run.sh, 5-run median, 2026-08-09). Node-based figures are vendor-published claims, not measured here. Coverage, not a superiority claim.