CLI Reference

July 23, 2026 · View on GitHub

SuperLocalMemory V3 Documentation https://superlocalmemory.com | Part of Qualixar

Complete reference for the slm command-line interface.

Run the interactive setup wizard. Package installation itself is non-interactive and does not install hooks, edit IDE configuration, start a daemon, or download a model. slm setup is the explicit activation boundary.

slm setup

`slm mode [a|b|c]`

Get or set the operating mode.

slm mode          # Show current mode
slm mode a        # Zero-cloud (no LLM, no API key)
slm mode b        # Local LLM via Ollama
slm mode c        # Cloud LLM (requires API key)

`slm provider [set]`

Get or set the LLM provider for Mode B/C.

slm provider          # Show current provider
slm provider set      # Interactive provider selector
slm provider set openai   # Set provider directly

`slm connect [ide]`

Configure IDE integrations.

slm connect           # Auto-detect and configure all IDEs
slm connect cursor    # Configure Cursor specifically
slm connect claude    # Configure Claude Code specifically

Run slm connect --list for the client names supported by the installed release. A documented configuration is not a claim that every client has passed the frozen V3.7 cross-client matrix.

Memory Operations

`slm remember "content" [options]`

Store a memory.

slm remember "API rate limit is 100 req/min on staging"
slm remember "Use camelCase for JS, snake_case for Python" --tags "style,convention"
slm remember "Maria owns the auth service" --scope shared --shared-with team-a
slm remember "Wait for all enrichment" --sync --json

Option	Description
`--tags "a,b"`	Comma-separated tags for categorization
`--sync`	Wait until declared derivation and projector stages complete
`--scope`	`personal`, `shared`, or `global` visibility
`--shared-with`	Comma-separated profile IDs for shared scope
`--json`	Emit the operation receipt and materialization state

Without --sync, the daemon returns after the memory is queryable; canonical enrichment continues in the background. If the daemon cannot start, the CLI stores raw evidence in the legacy offline spool. That spool is a compatibility input: replay submits the same source/idempotency identity through M018 before marking it done.

`slm recall "query" [options]`

Search your memories. Returns ranked evidence under Score Contract v2.

slm recall "rate limit"
slm recall "who owns auth" --limit 5

Option	Default	Description
`--limit N`	20	Maximum results to return
`--include-global`	off	Include global-scope facts when authorized
`--include-shared`	off	Include facts shared with the active profile

`slm search "query" [options]`

Alias for slm recall. Same behavior, same options.

`slm forget "query" [options]`

Delete memories matching a query.

slm forget "old staging credentials"             # Confirm before deletion
slm forget "old staging credentials" --dry-run   # Preview only
slm forget "old staging credentials" --yes       # Skip confirmation

Option	Description
`--dry-run`	Preview matches without deleting
`--yes`, `-y`	Skip the confirmation prompt
`--json`	Emit structured preview or mutation output

Use slm delete <fact_id> for precise deletion by ID.

`slm delete <fact_id>`

Delete a specific memory by its exact fact ID. Use slm list --json or slm recall --json to obtain fact IDs.

slm delete abc123
slm delete abc123 --yes    # Skip confirmation
slm delete abc123 --json

`slm update <fact_id> <content>`

Edit the content of a specific memory by its exact fact ID. The fact must belong to the active profile.

slm update abc123 "API rate limit is now 200 req/min on staging"
slm update abc123 "Updated content" --json

`slm list [options]`

List recent memories.

slm list              # Last 20 memories
slm list --limit 50   # Last 50 memories

V3 Features

`slm trace "query"`

Recall with a channel-by-channel breakdown. Shows how each retrieval channel contributed to the results.

slm trace "database port"

Output exposes the channels that contributed to each result. The current candidate producers are dense semantic, BM25 lexical, temporal, Hopfield associative, and spreading activation. Entity-graph information can enhance a post-fusion score but is not a separate candidate producer.

`slm health`

Show diagnostics for the mathematical layers.

slm health

Reports status of:

Fisher-Rao similarity layer
Sheaf consistency layer
Langevin lifecycle dynamics
Embedding model status
Database integrity

Migration

`slm migrate [options]`

Migrate a V2 database to V3 format.

slm migrate                # Run migration
slm migrate --rollback     # Roll back when a valid migration backup exists

Option	Description
`--rollback`	Revert to V2 format (backup must exist)

Use slm db migrate --dry-run to inspect additive database migrations.

Profile Management

`slm profile [command]`

Manage memory profiles (isolated memory contexts).

slm profile list                  # List all profiles
slm profile switch work           # Switch to "work" profile
slm profile create client-acme    # Create a new profile
slm profile list --json           # Structured profile inventory

System & Maintenance

`slm status`

Show system status: mode, profile, memory count, database location, health.

slm status

Global Options

Global metadata options are activation-free. Command-specific options appear in slm <command> --help; do not assume that an option accepted by one command is global.

Option	Description
`--help`	Show help for a command
`--version`	Show SLM version

Agent-Native JSON Output

Commands that advertise --json emit structured output. Recall results use Score Contract v2:

{
  "success": true,
  "command": "recall",
  "version": "<installed-version>",
  "data": {
    "results": [
      {
        "fact_id": "abc123",
        "content": "Database uses PostgreSQL 16",
        "relevance_score": 0.87,
        "ranking_score": 0.0132,
        "memory_confidence": 0.7,
        "rank_position": 1
      }
    ],
    "count": 1,
    "score_contract_version": "2",
    "calibration_status": "uncalibrated",
    "answer_confidence": null
  },
  "next_actions": [
    {"command": "slm list --json", "description": "List recent memories"}
  ]
}

Check slm <command> --help before scripting a surface; structured-output support is explicit per command and may expand between releases.

Usage with jq

# Get first result content
slm recall "auth" --json | jq '.data.results[0].content'

# Get all fact IDs
slm list --json | jq '.data.results[].fact_id'

# Check current mode
slm status --json | jq '.data.mode'

In CI/CD (GitHub Actions)

- name: Store deployment info
  run: slm remember "Deployed ${{ github.sha }} to production" --json

- name: Check memory health
  run: slm status --json | jq -e '.success'

Examples

Daily workflow

# Morning: check what you remembered yesterday
slm list --limit 10

# During work: store a decision
slm remember "Decided to use WebSocket instead of SSE for real-time updates" --tags "architecture"

# Later: recall the decision
slm recall "real-time communication approach"

# End of day: check system health
slm status

Project setup

# Create a profile for a new project
slm profile create mobile-app
slm profile switch mobile-app

# Store project context
slm remember "React Native 0.76 with Expo SDK 52"
slm remember "Backend is FastAPI on AWS ECS"
slm remember "CI/CD via GitHub Actions, deploys on merge to main"

Optimize Commands

SLM v3.6 adds the Optimize layer — Cache + Compress + Align for LLM cost reduction.

`slm optimize status|on|off|savings`

Master control for the Optimize module.

slm optimize status                # Show all settings
slm optimize on                    # Enable cache + compress
slm optimize off                   # Disable (proxy passes through)
slm optimize savings               # Token/cost report (last 7 days)
slm optimize savings --since 30    # Last 30 days
slm optimize savings --provider anthropic  # Per-provider filter
slm optimize savings --json

`slm cache status|clear|invalidate|ttl|semantic`

Cache sub-control. Exact cache is the stable path. Semantic cache remains experimental and must not be treated as enabled production behavior without a release-linked precision, invalidation, and tenant-isolation report.

slm cache status                   # Entry count, DB size, TTLs, hit rate
slm cache clear                    # Delete all entries (default tenant)
slm cache invalidate --tag "key"   # Delete entries by tag
slm cache ttl --set 86400          # Set exact TTL (24h default)
slm cache ttl --semantic 3600      # Set semantic TTL
slm cache semantic on|off          # Enable/disable semantic cache

`slm compress status|mode|prose`

Compression sub-control — active subcommands.

slm compress status                # Mode + per-channel state
slm compress mode safe|aggressive  # Set aggressiveness
slm compress prose on|off          # Prose compression (opt-in, aggressive mode only)

Removed in v3.6.10: slm compress code, slm compress ccr, and slm compress align no longer perform meaningful work and print a migration notice when invoked. Use slm compress prose on for prose-level compression.

`slm proxy [options]`

Start the optimization proxy (port 8765 by default).

slm proxy                          # Default port 8765
slm proxy --port 8080              # Custom port
slm proxy --provider anthropic     # Provider surface
slm proxy --no-compress            # Cache only
slm proxy --semantic               # Enable semantic cache

`slm wrap <agent> [options]`

Proxy-activate an agent — starts proxy + sets environment + launches agent.

slm wrap claude                    # Claude Code (recommended)
slm wrap cursor                    # Cursor
slm wrap aider -- --model gpt-4    # Aider
slm wrap --list                    # List registered agents
slm wrap --persistent              # Permanent config write
slm wrap --dry-run                 # Preview

`slm help-optimize [topic]`

Full developer reference with per-agent recipes.

slm help-optimize                  # Full reference
slm help-optimize cache            # Cache reference
slm help-optimize compress         # Compress + safety warning
slm help-optimize agents           # Per-agent setup
slm help-optimize safety           # Safety warning only

Full details: docs/optimize-cli.md

Daemon and System

`slm init`

One-command first-time setup: mode selection, hook installation, IDE configuration, and optional warmup. Equivalent to running slm setup then slm hooks install then slm connect then slm warmup interactively.

slm init           # Interactive wizard covering all setup steps
slm init --mode a  # Non-interactive: Mode A, no hooks, no IDE config

Option	Description
`--mode a/b/c`	Set operating mode non-interactively
`--no-hooks`	Skip hook installation
`--no-ide`	Skip IDE configuration

`slm doctor`

Pre-flight check: verifies dependencies, embedding worker, daemon connectivity, and configuration. Run after any install or upgrade.

slm doctor
slm doctor --json   # Structured output

`slm warmup`

Pre-download the embedding model (~500MB). Prevents the first-use download lag.

slm warmup

`slm dashboard`

Open the local web dashboard. Requires the daemon to be running.

slm dashboard           # Opens at http://localhost:8765
slm dashboard --port 9000

`slm serve [start|stop]`

Start or stop the background daemon. The daemon enables instant CLI response and is required by the dashboard, MCP server, and most commands.

slm serve start    # Start daemon (auto-started on most commands)
slm serve stop     # Stop daemon

`slm restart`

Restart the daemon — required after configuration changes that cannot take effect at runtime.

slm restart
slm restart --json

`slm mcp`

Start the MCP server in stdio transport mode. Used by IDE configurations that require a subprocess-based MCP server.

slm mcp

For HTTP transport (recommended), the daemon exposes /mcp/ automatically. See docs/mcp-tools.md.

`slm rotate-token`

Rotate the SLM install token. Run slm restart afterward to pick up the new token.

slm rotate-token

`slm disable [--reason "..."]`

Globally disable SLM. Writes a .disabled marker and stops the daemon. All subsequent commands that require the daemon return an informational message until slm enable is run.

slm disable
slm disable --reason "maintenance window"

`slm enable`

Remove the .disabled marker. Prints the command to start the daemon.

slm enable

`slm clear-cache`

Wipe regenerable caches. memory.db and learning.db are preserved — only derived caches and optimization state are cleared.

slm clear-cache

`slm reconfigure`

Re-run the interactive post-install configurator. Use to change the performance profile, operating mode, or deployment tier without a full reinstall.

slm reconfigure

`slm reap [--force] [--all]`

Find and optionally kill orphaned SLM processes.

slm reap            # Dry-run: list orphans
slm reap --force    # Kill orphaned processes
slm reap --all      # Kill ALL slm mcp processes (use after IDE switch)
slm reap --json

Hooks and IDE Integration

`slm hooks install|uninstall|status`

Manage auto-capture hooks for Claude Code and Codex.

slm hooks install              # Install for Claude Code (default)
slm hooks install --agent codex  # Install for Codex
slm hooks uninstall
slm hooks status

Option	Description
`--agent codex`	Target Codex instead of Claude Code
`--global`	Install in global config (all projects)
`--project`	Install in project-local config
`--dry-run`	Preview changes without writing

`slm codex install|remove|status`

Manage SLM add-ons for Codex: skills, subagents, and lifecycle hooks.

slm codex install       # Add SLM-owned skills, agents, hooks to Codex
slm codex remove        # Remove SLM-owned files from Codex
slm codex status        # Verify installation
slm codex install --dry-run

slm codex install is additive — it does not replace other agents' hooks or rewrite ~/.codex/config.toml. MCP wiring is a separate step: slm connect codex.

`slm connect [ide]`

Auto-configure an IDE integration. Detects the installed IDE and writes the appropriate MCP config snippet.

slm connect                # Auto-detect and configure all IDEs
slm connect claude-code    # Claude Code specifically
slm connect cursor         # Cursor
slm connect codex          # Codex MCP wiring
slm connect --list         # Show all supported IDEs

Sessions and Lifecycle

`slm session open|close`

Manage named sessions for context grouping and temporal summaries.

slm session open --project-path /path/to/project  # Warm context for a project
slm session open --query "auth service work"       # Explicit query
slm session close                                  # Close and summarize
slm session close --session-id abc123

`slm session-context [query]`

Print session context for use by hooks. Returns relevant memories for the current project path or an explicit query.

slm session-context                         # Auto-derive from cwd
slm session-context "auth service"          # Explicit query
slm session-context --max-results 5 --json

`slm observe [content]`

Submit content for automatic capture evaluation. The system decides whether the content contains a decision, bug fix, or preference worth storing.

echo "Decided to use WebSocket over SSE" | slm observe
slm observe "API rate limit is 100 req/min on staging"

V3.3 Lifecycle Commands

`slm decay [--execute]`

Run the Ebbinghaus forgetting decay cycle. Default is dry-run.

slm decay                   # Preview transitions
slm decay --execute         # Apply zone transitions
slm decay --profile work --execute
slm decay --json

`slm quantize [--execute]`

Run the EAP embedding quantization cycle. Default is dry-run.

slm quantize                # Preview changes
slm quantize --execute      # Apply precision changes
slm quantize --json

`slm consolidate [--cognitive]`

Run the memory consolidation pipeline.

slm consolidate                      # Standard consolidation
slm consolidate --cognitive          # Include CCQ cognitive consolidation
slm consolidate --dry-run            # Preview
slm consolidate --profile work
slm consolidate --json

`slm soft-prompts`

List active soft prompts — patterns the system has automatically learned from usage.

slm soft-prompts
slm soft-prompts --profile work --json

Data and Evidence

`slm evidence export|verify|import|rebuild`

Export, verify, import, or rebuild versioned memory evidence bundles. Evidence bundles are deterministic, checksummed JSONL archives.

slm evidence export /path/to/bundle.jsonl --profile default
slm evidence verify /path/to/bundle.jsonl
slm evidence import /path/to/bundle.jsonl --execute
slm evidence import /path/to/bundle.jsonl --replace --execute
slm evidence rebuild --execute   # Rebuild derived lexical state

Subcommand	Description
`export <dest>`	Write a deterministic checksummed JSONL bundle
`verify <bundle>`	Verify checksums and source reconciliation
`import <bundle>`	Import relational truth (dry-run unless `--execute`)
`rebuild`	Rebuild derived lexical state (dry-run unless `--execute`)

`slm diagnostics export <dest>`

Export bounded local operational aggregates. Produces a content-free JSON report (no memory content, no secrets) for manual inspection or support.

slm diagnostics export /path/to/report.json

`slm benchmark`

Run the evo-memory benchmark against an isolated temporary database. Never reads or writes user data.

slm benchmark
slm benchmark --json

Configuration and Adapters

`slm config get|set <key> [value]`

Get or set runtime configuration values using dot notation.

slm config get evolution.enabled
slm config set evolution.enabled true
slm config set security.require_login true
slm config set retention.default_policy gdpr-30d
slm config set embedding.model nomic-embed-text-v1.5

Common keys:

Key	Description
`evolution.enabled`	Enable/disable skill evolution
`security.require_login`	Require login for dashboard and API
`retention.default_policy`	`indefinite`, `gdpr-30d`, `hipaa-7y`, `custom`
`embedding.model`	Embedding model name
`mesh.discovery`	`on`/`off` for mDNS peer discovery

`slm adapters [subcommand]`

Manage ingestion adapters: Gmail, Calendar, and Transcript.

slm adapters list                  # List all adapters and their status
slm adapters enable gmail          # Enable Gmail adapter
slm adapters disable gmail         # Disable Gmail adapter
slm adapters start gmail           # Start adapter ingestion
slm adapters stop gmail            # Stop adapter ingestion
slm adapters status gmail          # Check adapter status

`slm ingest [--source ecc|jsonl]`

Import external observations into SLM's learning system.

slm ingest --source ecc            # Import Claude Code (ECC) sessions
slm ingest --source jsonl --file /path/to/data.jsonl
slm ingest --source ecc --dry-run  # Preview without writing

Supported sources:

ecc — Claude Code sessions (auto-discovers ECC observation files)
jsonl — Generic JSONL with content and optional timestamp fields

`slm evolve [--session <id>] [--profile <id>]`

Run post-session skill evolution. Normally called automatically by the Stop hook; invoke manually to process a specific session.

slm evolve
slm evolve --session abc123 --profile work

Bounded Loops (v3.8.0)

`slm loop demo [--iterations N] [--json]`

Run the keyless convergence demo. A stub proposer runs laps against a deterministic gate (fails twice, passes on lap 3). Every lap is written to SLM memory under tag loop:convergence-demo and is visible on the dashboard.

slm loop demo
slm loop demo --iterations 5
slm loop demo --json

Confirms the bounded-loop engine, SLM ledger, and gate mechanism end to end without a credentialed agent or a running daemon. Use it to verify the installation after an upgrade.

`slm loop history [--name NAME] [--json]`

List recorded runs for a loop name from the SLM ledger.

slm loop history                          # defaults to "convergence-demo"
slm loop history --name deploy-gate
slm loop history --name deploy-gate --json

Option	Default	Description
`--name NAME`	`convergence-demo`	Loop name to query
`--json`	off	Emit structured JSON

`slm loop show <run_id> [--json]`

Show every lap of one run, in order.

slm loop show <run_id>
slm loop show <run_id> --json

run_id is printed by slm loop demo and returned by the slm_loop_run MCP tool. Each lap row records decision, gate pass/fail, detail text, and token budget.

The MCP equivalents are slm_loop_run, slm_loop_history, and slm_loop_show. See MCP Tools Reference → Bounded-Loop Tools.

Database Maintenance

`slm db migrate [options]`

Inspect or run additive database schema migrations.

slm db migrate             # Apply pending migrations
slm db migrate --dry-run   # Preview pending migrations
slm db migrate --rollback  # Roll back to previous schema backup

`slm db scale status|prepare|verify|promote|rollback|adopt`

Manage the optional Scale Engine (CozoDB graph + LanceDB vector projections).

slm db scale status                              # Show current Scale Engine state
slm db scale prepare                             # Stage a new projection
slm db scale verify --stage-id <id>             # Verify parity with canonical SQLite
slm db scale promote --stage-id <id>            # Promote verified projection
slm db scale rollback --backup-id <id>          # Roll back to a prior projection
slm db scale adopt                               # Adopt a detected pre-v3.7 projection

SQLite + sqlite-vec remain canonical. Projections are parity-gated; a failed verify leaves recall on SQLite with the rejected manifest retained for inspection.