CLI Reference

Name: Savants
Author: Savants

The savants CLI is a single binary with 37 MCP tools for code intelligence, guard management, and infrastructure monitoring. Zero dependencies — no bash, no Python, no runtime needed. Works on macOS, Linux (including NixOS), and Windows.

CLI vs. npm package: The CLI gives your LLM code intelligence tools (semantic search, error diagnosis, blast radius). The npm package (@savants.dev/guard) adds guardrails to your AI agent code. You can use both together or independently.

Install

# macOS / Linux
curl -fsSL savants.sh | sh

# Windows (PowerShell)
irm releases.savants.dev/latest/install.ps1 | iex

Code Intelligence

These tools help your LLM understand your codebase. They run as an MCP server for Claude Code, Cursor, and Windsurf.

Command	Description
`savants up`	Start the MCP server (auto-detects repos, clusters)
`savants mcp install`	Register MCP tools with your IDE
`savants diagnose <error>`	Trace error to root cause with source context
`savants search <query>`	Semantic code search across all indexed repos
`savants callers <function>`	Find all callers of a function (recursive)
`savants blast-radius <file>`	Show impact of changing a file

Profile Commands

Command	Description
`savants profile install <slug>`	Install a profile from the registry
`savants profile search <query>`	Search the registry
`savants profile publish <dir>`	Publish your profile to the registry
`savants profile info <slug>`	Show profile details

Install a profile

savants profile install @savants/ai-agent-guardrails

# Output:
#   ✓ Schema loaded: agent-action (6 states, 7 transitions)
#   ✓ Guard set: block destructive actions in production
#   ✓ Guard set: require approval for send actions
#   ✓ Profile applied successfully

Guard Commands

Command	Description
`savants guard add "<rule>"`	Add a guard rule in DSL
`savants guard list`	List all active guards
`savants guard test "<rule>" --context '...'`	Test a guard with sample data
`savants guard remove <id>`	Remove a guard
`savants guard share <name> --version <ver>`	Publish a guard profile to the cloud
`savants guard browse`	Browse popular guard profiles
`savants guard update [name]`	Update installed profiles (or `--check` for dry run)
`savants guard pin <name> <version>`	Pin a profile to an exact version
`savants guard versions <name>`	List all available versions of a profile
`savants guard rollback <name>`	Roll back a profile to its previous version
`savants guard why <rule-id>`	Explain why a guard rule exists and when it fires
`savants guard sync push`	Push local guard rules to the cloud
`savants guard sync pull`	Pull guard rules from the cloud to this machine

Add a guard

savants guard add "when action contains 'delete' and env eq 'production' then block"

Test a guard

savants guard test "when spend gt 100 then block" --context '{"spend": 150}'

# Output:
#   ✓ Guard FIRED → action: block
#   Context: spend=150, threshold=100

Monitor Commands

Command	Description
`savants monitor start`	Start infrastructure monitoring
`savants monitor status`	Show current entity states
`savants monitor history <entity>`	Show transition log for an entity

General Commands

Command	Description
`savants update`	Update Savants to the latest version
`savants connect`	Connect to Savants Cloud
`savants reindex`	Re-index the current codebase

Configuration

Telemetry

Savants collects privacy-respecting, anonymous usage telemetry to improve the product. No code, file paths, or personal data is ever sent. You can opt out at any time:

# Check current telemetry status
savants config telemetry status

# Opt out of telemetry
savants config telemetry off

# Opt back in
savants config telemetry on

You can also set the environment variable DO_NOT_TRACK=1 to disable telemetry without running a command.

Uninstall

# macOS / Linux
curl -fsSL releases.savants.dev/latest/uninstall.sh | sh

# Windows (PowerShell)
irm releases.savants.dev/latest/uninstall.ps1 | iex