sealevel 0.3.0

CLI for Sealevel — the Solana-specialized coding LLM. Chat, generate Anchor programs, review code, decode errors.

Sealevel CLI

Command-line interface for Sealevel — the Solana-specialized coding AI.

Interactive terminal session with agent mode, slash commands, and Solana/Anchor developer tools.

Install

pip install sealevel

Requires Python 3.10+.

Quick Start

slm login              # Authenticate via browser (OAuth device flow)
slm                    # Start interactive session

Or set your key manually:

slm config --api-key slm_xxxxxxxxxxxx

Interactive Session

❯ How do I derive a PDA in Anchor?
◆ SEALEVEL
[streaming response with markdown rendering]

❯ /review @src/lib.rs
◆ REVIEWING  src/lib.rs
[security + deprecated pattern analysis]

❯ /explain-error 0x1771
[error decode + fix suggestion]

❯ /gen escrow with atomic token swap -o src/lib.rs
✓ WROTE  src/lib.rs

Agent Mode

Toggle with /agent — the LLM can read, edit, and create files, and run commands:

❯ /agent
✓ Agent mode ON

❯ read src/lib.rs and add an authority check to increment
╭─ 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

Done. Added authority check.

Tools: read_file, write_file, edit_file, run_command, glob_files, grep_files. Read-only tools auto-approve. Write/execute asks permission. Type a to approve all.

Features

• @file references — @src/lib.rs inlines file content into chat
• Project memory — SEALEVEL.md in project root auto-injects context
• AI-powered /compact — summarizes old conversation to free context
• Auto-upgrade — notifies when newer version is available
• File checkpoints — /undo restores files modified by agent
• Persistent history — input history persists across sessions (Ctrl+R to search)
• Progressive streaming — live markdown rendering with Rich

Modes

slm                    # Interactive session (default)
slm -p "prompt"        # One-shot pipe mode
slm -c                 # Continue most recent session
slm login              # Authenticate via browser
slm logout             # Clear credentials
slm config --show      # View configuration

Pipe mode

slm -p "What is a PDA?"
cat src/lib.rs | slm -p "review this code"

Slash Commands

Type / for live dropdown. 25 commands:

Code

Command	Description
/review <file>	Security + deprecation review
/migrate <file> [--write]	Migrate to Anchor 0.30+
/gen <description> [-o file]	Generate Anchor program
/tests <file> [-o out.ts]	Generate TypeScript tests

Explain

Command	Description
/explain-tx <signature>	Decode a Solana transaction
/explain-error <code>	Decode an error code

Session

Command	Description
/sessions	List past sessions
/resume <id>	Resume a past session
/rename <name>	Rename current session
/history	Show conversation history
/search <query>	Search conversation
/compact [focus]	AI-summarize old history
/export [file]	Export as markdown
/clear	Clear history (with confirmation)
/undo	Undo last turn + restore files
/retry	Redo last turn

Info

Command	Description
/status	API health + config
/usage	Token usage + limits
/copy	Copy last response to clipboard

System

Command	Description
/agent	Toggle agent mode
/login	Authenticate via browser
/config [--show]	View/set config
/rotate-key	Rotate API key
/help	Show all commands
/exit	Exit session

Configuration

slm config --api-key slm_xxxx     # Set API key
slm config --api-url https://...  # Custom endpoint
slm config --mode quality         # 'quality' (temp=0.0, 4096 tokens) or 'fast' (temp=0.3, 2048)
slm config --show                 # Show current config

Config: ~/.sealevel/config.toml (chmod 600). API key: OS keyring.

License

MIT

Links

• Sealevel
• Docs
• PyPI