CLI — Sealevel Docs

Install

Bash

pip install sealevel
# Requires Python 3.10+

Authenticate

Bash

slm login          # Opens browser, authenticates via OAuth device flow
slm logout         # Clear stored credentials
slm config --show  # Verify your key is set

Your API key is stored securely in the OS keyring. You can also set it manually with slm config --api-key slm_YOUR_KEY.

Interactive session

Run slm with no arguments to start an interactive session. Type plain text to chat, or use slash commands:

Bash

$ slm
❯ How do I derive a PDA in Anchor?
◆ SEALEVEL
  To derive a PDA, use seeds and bump in your #[account] attribute...

❯ /review src/lib.rs
◆ REVIEWING  src/lib.rs
  ...

❯ /gen counter with increment and decrement -o src/lib.rs
✓ WROTE  src/lib.rs

Slash commands

Type / to see the live dropdown. 25 commands:

Code
/review <file> — Security + deprecation review for Solana/Anchor code
/migrate <file> [--write] — Upgrade to modern Anchor 0.30+ patterns
/gen <description> [-o file] — Generate a complete Anchor program
/tests <file> [-o out.ts] — Generate TypeScript tests
Explain
/explain-tx <signature> — Decode a Solana transaction
/explain-error <code> — Decode a Solana/Anchor error code
Session
/sessions — List past sessions
/resume <id> — Resume a past session
/rename <name> — Rename current session
/history — Show conversation history
/search <query> — Search conversation history
/compact [focus] — AI-summarize old history to free context
/export [file] — Export session as markdown
/clear — Clear conversation history (with confirmation)
/undo — Undo last turn + restore modified files
/retry — Redo last turn with a fresh response
Info
/status — API health + config overview
/usage — Token usage and limits
/copy — Copy last response to clipboard
System
/agent — Toggle agent mode — experimental
/login — Authenticate via browser
/config [--show] — View or change settings
/rotate-key — Rotate API key
/help — Show all commands
/exit — Exit the session

Agent mode (experimental)

Toggle with /agent. The model can read files, search code, and run commands. Best with one action at a time:

Bash

❯ /agent
✓ Agent mode ON

❯ read src/lib.rs and add an authority check
╭─ read_file ──────────────╮
│  path: src/lib.rs        │
│  ✓ 38 lines              │
╰──────────────────────────╯
╭─ edit_file ──────────────╮
│  path: src/lib.rs        │
│  -2 / +5 lines           │
╰──────────────────────────╯
▸ Allow edit? [y/N/a] y
✓ EDITED  src/lib.rs

Read-only tools auto-approve. Write and execute tools ask permission. Type a to approve all of that type for the session. Use /undo to revert file changes.

Inline file references

Use @path/to/file.rs in chat to include file contents inline:

Bash

❯ What's wrong with @src/lib.rs?
# File content is automatically injected into the prompt

Project memory

Create a SEALEVEL.md file in your project root (or ~/.sealevel/SEALEVEL.md) to inject project-specific context into every prompt. The CLI walks from cwd to root looking for it.

Pipe mode

For scripting and CI, use -p for one-shot prompts:

Bash

slm -p "What is a PDA?"                           # one-shot
cat src/lib.rs | slm -p "review this code"        # pipe stdin
slm -c                                            # continue last session

First-run onboarding

On first launch without an API key, the CLI prompts you to authenticate:

Bash

$ slm
◆ Sealevel  v0.3.0  ·  sealevel.tech

Sign in to get started:

  1. Browser login (recommended)
  2. Paste API key manually

Choose [1/2]:

Option 1 runs the device flow (same as slm login). Option 2 prompts for a key with masked input. You can skip and authenticate later with /login inside the session.

Modes

Switch between quality and fast inference:

Bash

slm config --mode quality   # temp=0.0, 4096 max tokens (default)
slm config --mode fast      # temp=0.3, 2048 max tokens

Shell completion

Bash

slm --install-completion bash   # or zsh | fish