CLI commands

Flags

Keyboard shortcuts

Slash commands

Navigation

Files & Code

Git

Memory & Sessions

Analysis & Planning

Settings

Modes of operation

Interactive TUI (default)

atlas
atlas --model atlas-plan
atlas --no-agent     # plain chat, no tool use
atlas --auto-model  # auto-select model by task type

One-shot mode

Prints the answer directly to stdout. No TUI. Good for scripts.

atlas ask "what does auth.ts do?"
atlas ask "summarise the last 5 commits" --model atlas-plan

Pipe mode

When stdin is not a TTY, Atlas reads it and combines with your prompt.

cat error.log | atlas "why is this failing?"
git diff | atlas "write a commit message"
cat src/auth.ts | atlas   # analyse without a prompt

Workspace setup

Run atlas init in your project root to create the workspace:

.atlas/
  memory.md        ← session memory (auto-updated every 3 turns)
  permissions.json ← permission config
  commands/        ← custom slash commands (*.md files)
  backups/         ← file backups before each edit
  audit.jsonl      ← tool action log
ATLAS.md           ← project context injected every session

Example ATLAS.md:

# My App

Next.js 14 · TypeScript strict · PostgreSQL · Prisma ORM

## Rules
- Always run `npm test` after making changes
- Never modify files in /generated/ or /migrations/
- Use Zod for all runtime validation
- Components go in src/components/, hooks in src/hooks/

Memory system

Atlas automatically builds and maintains .atlas/memory.md as you work.

Session restore

Full conversations are saved to ~/.atlas/sessions/ after every turn. Sessions older than 30 days are pruned automatically.

# List saved sessions
/history

Saved sessions (newest first):
  1. [11 May 2026, 17:45]  dev/myapp · 12 turns
     "add JWT authentication to the Express API…"
  2. [10 May 2026, 09:30]  dev/myapp · 4 turns
     "refactor the database connection pool…"

# Restore session 1
/resume 1

Restoring replays the last exchange so you immediately see the context, then continues from the full message history.

Fix loop

The /fix command creates an automated repair cycle:

# Auto-detect test command
/fix

# Specific command
/fix cargo test
/fix pytest -x
/fix npm run lint

How it works:

🔧 Attempt 1/5 — running: npm test
✗ Failed (exit 1) — asking Atlas to fix…

  ✓ ◎ read  src/utils/parser.ts
  ✓ ✎ edit  src/utils/parser.ts → Fixed regex on line 42

🔧 Attempt 2/5 — running: npm test
✓ All checks pass after 2 fixes!

Supports auto-detection for: npm test, npm run build, cargo test, go test ./..., pytest. Any custom command also works.

Auto-compact

When the conversation reaches ~80K tokens, Atlas automatically:

📦 Context ~82K tokens — auto-compacting…
✓ Compacted — continuing with summary + last 6 messages

The compaction writes a dense 400-word summary of all decisions and changes, replaces the conversation history with it, and continues seamlessly. If compaction fails for any reason, Atlas proceeds with the full context.

You can also compact manually at any time with /compact.

Custom slash commands

Create .atlas/commands/<name>.md to define your own commands. The file content is the prompt sent to Atlas. Use $ARGUMENTS for anything typed after the command name.

# .atlas/commands/deploy.md
Run the deployment checklist for $ARGUMENTS:
1. Run `npm test` — abort if any fail
2. Run `npm run build`
3. Check git status — all changes committed?
4. Output the deploy command for $ARGUMENTS

# Usage:
/deploy staging
/deploy production

Configuration

Config is stored in ~/.atlas/config.json.

atlas config set apiKey sk-atlas-xxxx
atlas config set baseUrl https://atlas.alpheric.ai/v1
atlas config set model atlas-plan