Runs Claude Code under the hood  ·  macOS

Go get a coffee.
Your features
are building themselves.

For people who are done babysitting Claude across a dozen tabs. Set context. Walk away. Come back to done.

See how it works  →
TerminalAI — changelog being generated with Explore panel open
BA
AH
SR
ML
Used by creative developers, directors who code, and solo operators running too many projects at once

"I was feeling like a monkey pressing Enter, keeping tabs on 12 sessions in various states of chaos. Terminal AI is the Ferrari version of what's already a thousand-X product."

Brandon Aviram — built it to solve his own problem
Under the hood

Claude Code.
With a cockpit.

Terminal AI doesn't reinvent the engine. It runs Claude Code — Anthropic's own agentic coding CLI — underneath everything. You're not betting on something unproven. You get the full power of Claude Code plus the interface and workflow layer that makes it actually livable.

Your existing account. Your existing setup. Terminal AI just gets out of the way and lets it work.

You Describe what you want. Walk away.
↓
Terminal AI Interface  ·  Skills  ·  Context triage  ·  Sessions  ·  Agent panel
↓
cc/
Claude Code Anthropic's agentic engine. The actual execution.
cx/
Codex OpenAI's coding agent. Switch providers.
Why this matters
✓
No new trust required
Claude Code is Anthropic's own tool — battle-tested, actively developed. Terminal AI inherits all of that credibility.
✓
Use your existing account
No new API keys. No new billing. Terminal AI runs through your existing Claude Code or Codex setup. Bring your own accounts — install once, run everything.
✓
The interface Claude Code never had
Skills, Focus dashboard, live agent progress, Side Quest, persistent sessions — all the workflow Claude Code leaves up to you.
✓
Stays current automatically
When Anthropic improves Claude Code, you get it. Terminal AI rides the wave — you never fall behind the model.
Features

Every tool you need.
Nothing you don't.

The best practices you'd normally keep in markdown files and CLAUDE.md configs — baked in. You shouldn't have to build a tmux setup just to stay organized.

⚡
Parallel Agent Runs
Spawn multiple agents simultaneously. Watch them tackle your codebase from different angles — then come back to done.
live agent panel
🧠
Skill System
Teach Terminal AI your conventions and preferences once. It applies them automatically on every run — forever.
~/.claude/skills/
🎯
Focus Dashboard
Surfaces your most relevant project based on recent activity and git sync status. The right context, always at the top.
garden · focus · classic
🔍
Side Quest
A separate scratch window for quick questions that won't pollute your main context. Ask, get an answer, keep going.
⇧⌘E
💾
Persistent Sessions
Pick up exactly where you left off. Conversations persist across restarts — no re-reading 500 lines of terminal scroll.
git-friendly · searchable
⚓
Keyboard-First Flow
⌘K palette, ⌘T new tab, J/K navigation. Every action is one chord away. Your hands never leave the keyboard.
⌘K · vi nav
How it works

Set it.
Walk away.
Come back to done.

One core loop: you provide intent, agents provide execution. You don't have to watch the sausage get made.

01
Open your project
Terminal AI reads your project structure, git history, and loaded skills to build a complete picture of your codebase — instantly.
02
Describe what you want
Type or paste a screenshot. Terminal AI parses your intent and plans the work before writing a single line of code.
03
Agents get to work
Parallel agents tackle the problem from multiple angles. Watch the agent panel — or close the laptop. Come back to done.
04
Pick up where it left off
Sessions persist. The Explore sidebar shows every tool call. The Devlog keeps the full history. Nothing gets lost between coffee runs.
Agent Run  ·  share-links ● 2 active
diagnose:nav
0:42
impl:sw-cache
1:18
write:tests
0:51
deploy:build
—
Ship Log

What we shipped

May 02, 2026
Latest
New Powers
  • Read the Explore panel like a story: ask, actions, landing
  • Scan branch and session count at a glance on the Airy dashboard
  • See each project's current state without opening it
  • Stop re-explaining the same lesson — TerminalAI remembers across sessions
  • Browse what TerminalAI learned and undo any lesson with one key
  • Resolve conflicting learnings inline without leaving the pod
  • Tighter About window and conversation actions
More Reliable
  • Drop screenshots and Finder copies straight into chat
  • See "Thinking" while Claude works, not "Listening"
  • Hit Return in side query; watch tools run live
  • Retry chatter stays quiet when commands fail
  • Toolbar reclaims its space when Verify is hidden
  • Side queries survive the orphan-process sweep
  • Run parallel prompts without crashing learning extraction
Under the Hood
  • Stop losing hard-won lessons between sessions
  • Skip API key setup — claude auth carries through
  • Avoid the silent ClaudeClient.chat failure mode
  • Sharper positioning grounded in actual customer language
  • Landing page now speaks the way buyers think
Apr 19, 2026 5 improvements →
New Powers
  • See agents finish even when Claude goes quiet — auto-recap card.
  • Track total spend across the conversation, not just last reply.
  • Stop seeing false "stalled" alarms during long tools and thinking.
  • Run builds and deploys without the stall warning lighting up.
Removed
  • Stop hunting for a button that just toggled notifications.
Apr 01, 2026 16 improvements →
New Powers
  • See exactly when the next retry fires, no more blank stares
  • Smart backoff with jitter replaces dumb 2s/4s retry delays
  • Wait out rate limits instead of slamming a closed window
  • Pick up where you left off — SessionCompactor rebuilds real context, not raw dumps
  • Watch agent budgets live in each row — "Step 12/50" with amber warnings
  • Stop hammering the API near rate limits — auto-throttles past 80%
More Reliable
  • Stop seeing blank replies when tools run but nothing comes back
  • Get all 3 retry attempts instead of stalling at 2
  • Cancel a stuck request without breaking the next one
  • No more phantom slowdowns from over-fired concurrent requests
  • Wait out rate limits instead of burning retries on a closed window
  • Large directory trees no longer freeze the response
Under the Hood
  • RetryPolicy and SessionCompactor land as standalone modules
  • SessionPhase.retrying carries attempt, reason, and reset time without animation churn
  • OpenCode long-run patterns documented in docs/opencode-agent-longrun-patterns.md
  • Independent agent and Codex audits cleared every retry path
Mar 29, 2026 16 improvements →
New Powers
  • Six live status states — stop guessing if your agent stalled
  • Stalled agents auto-recover in 60s — no babysitting required
  • Explore panel follows tool activity — stop scrolling to catch up
  • Switch branches in terminal — toolbar updates instantly
  • Linear, Calendar, Grain connect without permission errors
  • Resume old sessions with full context, not a blank slate
More Reliable
  • Finishing... never freezes — responses complete even when CLI drops
  • Cancel kills hanging calls instantly, no leaked background tasks
  • Expired sessions auto-recover instead of showing blank bubbles
  • Claude stays in your project folder, stops wandering elsewhere
  • Status bar waits until it matters, dismisses itself on all content
  • Cleaner context injection — stray paths no longer confuse Claude
  • Dashboard icon opens new tab, never hijacks your current chat
Under the Hood
  • Simpler watchdog — fewer places stall detection can break
  • Streaming never waits on telemetry writes
  • Stalls, failures, wrong_cwd — four blind spots now visible
Mar 26, 2026 12 improvements →
New Powers
  • Switch AI backends from settings — no restart, no env vars
  • Markdown tables render clean — no more layout seizures
  • ⌘K shows only commands that matter right now
  • Dead sessions resume in 15s, not 30
  • Reopen finished chats instantly — skip the stale handshake
  • Stop accidentally spawning parallel streams
More Reliable
  • Subagent names render clean — no stray { artifacts
  • Cancel parallel agents without leftover ghost spinners
  • Scan tool output at a glance — real labels, not raw JSON
  • Agent SDK finds node on first launch — zero path debugging
Under the Hood
  • Zero manual steps — dev.sh auto-installs agent bridge deps
  • /tai — stop typing /terminal-ai-dev every session
Mar 23, 2026 26 improvements →
What Changed (builds 95–107)
  • **Conversation Style picker renamed**: "Document" → "Artifact" across the whole codebase
  • **Gradient separators** between messages in Artifact mode — premium fade instead of flat hairline
  • **Dashboard fix**: ⌘D / Escape now opens a new tab rather than replacing the current chat view (active chat no longer disappears)
  • New **Conversation Style** toggle in the toolbar: **Uniform** (standard) and **Artifact** (document-style with larger type, more breathing room, named separators)
  • Layout, spacing, and naming refined through builds 105–106
  • Image drop detection expanded — NSTextView paths and additional UTTypes now handled correctly
  • All disclosure animations across the app replaced with spring physics (response: 0.22, dampingFraction: 0.82) — feels native and instant
  • Scale-from-anchor transitions (0.97) replace .move(edge:) slides app-wide
  • Explore panel chevrons rotate with spring; expand/collapse blocks feel snappy
  • Tooltip content pruned — Explore rows show label only, not a raw content dump
  • **Drag-drop reliability** — all NSItemProvider return types handled; files that previously silently failed now load correctly
  • **Tooltip verbosity** — silenced harsh Explore panel tooltips that were dumping raw tool content
  • **Dashboard tab behavior** — return to dashboard no longer destroys the active chat
What Changed (build 108)
  • Headings register at a glance — H1/H2/H3 no longer blur together
  • Conversations breathe — 40px between exchanges reads like a document
  • CLAUDE label hugs its response instead of floating in dead space
  • Claude's replies land with visible weight, not lost in the scroll
  • Paragraph rhythm tuned for reading, not chat-app skimming
What Changed (build 109)
  • Dead-session blank stare cut in half — recovery fires at 15s
  • Status text stays ahead of the timeout, no UI flicker
  • _Analyzing context..._ shows in ~15s, not ~30s
What Changed (build 110)
  • Stop button never disappears mid-processing
  • Save recordings without a beach ball
  • App starts without freezing on session decode
  • Hung CLI fails at 45s, not two minutes
  • Reopen finished sessions without the 29s --resume hang
Efficiency

Master the keyboard

Command palette
⌘K
New chat
⌘N
New tab (instant)
⌘T
Side Quest
⇧⌘E
Return to dashboard
⌘D
Cancel / close
ESC
Checkpoint session
⇧⌘S
Close all panels
⇧⌘W

Stop babysitting.
Start shipping.

macOS 14+  ·  Requires Claude Code CLI  ·  Uses your existing account

No spam. You'll hear when your spot opens.