CLI Command Reference¶

All OLAV features are accessible via the command line.

Feature Claims

ID	Claim	Status
C-L2-01	`olav version`	✅ v0.10.0
C-L2-02	`olav list`	✅ v0.10.0
C-L2-10	`olav export claude-plugin`	✅ v0.10.0
C-L2-13	`olav init`	✅ v0.10.0
C-L2-27	`olav reset --agent`	✅ v0.10.0
C-L2-14	TUI interactive mode	✅ v0.10.0

Basic Usage¶

olav [options] [command] [query]

Global Options¶

Option	Short	Default	Description
`--agent NAME`	`-a`	`core`	Specify which Agent to use
`--auto-approve`		off	Automatically approve all tool calls (skip confirmation)
`--dangerously-skip-permissions`		off	Skip permission checks (development only)
`--sandbox MODE`		`none`	Sandbox mode: `none` (local) / `modal` / `daytona` / `runloop` (remote)
`--verbose`	`-v`	off	Show verbose output (for debugging)
`--no-splash`		off	Skip startup banner
`--session ID`			Resume a previous session
`--version`	`-V`		Show version number

Command List¶

Basic Commands¶

Command	Description
`olav "your question"`	Natural language query (using the default Agent)
`olav --agent <name> "..."`	Query with a specific Agent (v0.18.1 top-level: `core` / `audit` / `netops` / `services`)
`olav`	Launch the interactive terminal (TUI)
`olav version`	Show the platform version
`olav list`	List all available Agents
`olav init`	Initialize an OLAV project in the current directory

Workspace Management¶

Command	Description
`olav workspace list`	List all workspaces (`*` marks the active one)
`olav workspace use <name>`	Switch the active workspace
`olav workspace status`	Show workspace status

Agent / Skill Management¶

Command	Description
`olav agent install <path\\|URL>`	Install an agent extension package (workspace + tools) from a local directory, wheel, or Git repository
`olav agent install <path> --merge-into <ws>`	Append tools to an existing workspace instead of registering a new agent
`olav skills list`	List installed skills (skills are capability bundles inside an agent)
`olav skills info <name>`	View skill details (version, source, install time)

v0.20.3 command rename

olav skill install <path> (singular skill) was removed in v0.20.3; use olav agent install <path> instead. v0.21.0-rc4 ships a backward-compatibility shim: running the old command prints a yellow warning and forwards to the new one, so behaviour is identical. The plural olav skills verb (managing the skill catalogue) is a separate command and is unaffected.

Migration:

# Old (v0.19):
olav skill install /path/to/olav-netops/

# New (v0.20.3+):
olav agent install /path/to/olav-netops/

# Both do the same: auto `pip install` the package's dependencies,
# deploy its workspaces, and rebuild the router index
# (no separate `olav refresh` needed).

Service Registry¶

Command	Description
`olav registry register <name>`	Register an external service (auto-discovers OpenAPI schema and generates tools)
`olav registry refresh <name>`	Force re-fetch the schema of a registered service
`olav registry list`	List all registered external services
`olav registry status <name>`	Check the reachability of a registered service

Audit and Logs¶

Command	Description
`olav log list`	Operations from the last 24 hours
`olav log errors [--hours N]`	Show only error events
`olav log show <run-id>`	View full details of a specific operation
`olav log export sft\\|trajectory\\|atif`	Export audit data in training formats

Background Services¶

Command	Description
`olav service web start [--port N]`	Start the Web service (default port 2280)
`olav service daemon start`	Start the Daemon accelerator service
`olav service logs start [--port N]`	Start the Syslog receiver service (default port 5514)
`olav service start --all`	Start all background services
`olav service stop --all`	Stop all background services
`olav service status`	View service status

User Management¶

OLAV users are aligned to Linux system users — olav user create alice requires pwd.getpwnam('alice') to succeed first. No parallel identity store. Per-user encrypted tokens (sha256 hash + salt in users.duckdb, plaintext shown once at create) authenticate against the web API.

Command	Description
`olav user create <name> [--role admin\\|user\\|readonly]`	Issue token for an existing Linux user. `--no-verify` to skip the `getpwnam` check (container deployments).
`olav user list`	Tabular roster of OLAV-registered users
`olav user token <name>`	Rotate + print a fresh token (revokes the old hash)
`olav user revoke <name>`	Disable the user without deleting the row

Legacy verbose form (olav admin add-user / list-users / rotate-token / revoke-token) is still supported and dispatched to the same implementation.

Knowledge Base Management¶

Command	Description
`olav kb import-guides [<dir>]`	Prime every `*.guide.yaml` under a workspace into LanceDB (idempotent; orphan-prune removes rows whose source YAML was deleted)
`olav kb list-guides [--tier vendor\\|platform\\|team\\|user] [--agent <name>]`	Tabular view of active guides + their `source_tier`
`olav kb show <intent> [--agent <name>]`	Print the YAML source (use for promotion: copy → git PR → merge → re-prime)
`olav kb remove <intent> --reason "<text>" [--agent <name>] [--keep-yaml]`	Tombstone the YAML (`<intent>.guide.yaml.removed`) + delete LanceDB row + drop a `kb_audit/<ts>_remove_<intent>.yaml` row. `--reason` required (escape: `OLAV_KB_REMOVE_ALLOW_NO_REASON=1`)
`olav kb status`	Knowledge store statistics
`olav kb export [--dir <path>]`	Export vault to Obsidian markdown
`olav kb search <query>`	Search the unified knowledge store

Schema Evolution¶

Command	Description
`olav config evolve --list`	View pending schema evolution proposals
`olav config evolve --approve <id>`	Approve and apply a schema evolution

Other¶

Command	Description
`olav export claude-plugin`	Export Claude plugin artifacts
`olav reset --agent <name>`	Reset an Agent's conversation history

Diagnostic & Discovery Commands¶

Round 25 (ARCH-12), Round 45 (ARCH-11 explainability), and Round 47 (ARCH-13) landed three user-level shortcuts that talk to the data surface without running the agent loop.

Command	Description
`olav catalog`	List the data-model topics (Device Inventory / Routing & BGP / Topology / …).
`olav catalog <topic>`	Drill into a topic: lists the underlying tables + example queries.
`olav catalog describe <table>`	Full schema for a table — columns, recipe mappings, samples.
`olav explain "<token>"`	Resolve a citation token from an audit report (e.g. `[src: netops.parsed_outputs#snap_a; device=R1; row=4]`) back to the raw DuckDB row(s). Read-only.
`olav diff <snap1> <snap2>`	Cross-snapshot diff aggregator over parsed_outputs / topology_links / raw_output_store / oc_outputs. Flags: `--table`, `--device`, `--max-rows`, `--json`.

Examples:

# What can I query about BGP?
olav catalog "Routing & BGP"
olav catalog describe netops.parsed_outputs

# Verify an audit finding's source row
olav explain "[src: netops.parsed_outputs#snap_20260418_0201; device=R1; row=4]"

# What changed between two snapshots?
olav diff snap_20260418_0100 latest --device R1 --max-rows 10
olav diff snap_a snap_b --table parsed_outputs --json

All three commands are auth-gated but read-only; they don't invoke an LLM or modify state.

Interactive Mode (TUI)¶

Run olav without a query argument to launch the interactive terminal:

olav                      # Use the default Agent (core)
olav --agent audit        # Use a specific Agent (e.g. audit / netops / services)
olav --no-splash          # Skip the startup banner
olav --session <id>       # Resume a previous session

Slash Commands in Interactive Mode¶

Built-in CLI verbs¶

Command	Description
`/help`	Show the list of available commands
`/clear`	Reset the conversation and clear the screen
`/model <name>`	Temporarily switch the LLM model within the session (does not modify config files)
`/trace-review`	Analyze failures from the last 7 days and automatically extract lessons learned into memory
`/quit` `/exit` `/q`	Exit interactive mode

Agent-scoped slash commands¶

These commands are deterministic Python wrappers shipped by specific agents. Invoke as olav --agent <name> "/<command>" or via /<command> inside an interactive session running that agent.

Slash	Agent	Description
`/netops_init`	`netops`	Full discovery pipeline — SSH all devices, parse, ETL into DuckDB, build per-command auto-views
`/export_netbox_csv [--filename X] [--tenant T]`	`netops`	Export `netops.devices` to a NetBox-import-compatible CSV under `exports/`. See the connect-a-service guide for the round-trip workflow.
`/import_netbox_csv [--csv X] [--report Y] [--write --endpoint URL --token TOK]`	`services`	Validate (default) or push (with `--write`) a NetBox-shaped CSV. Idempotent 5-step lookup-or-create chain. Drops a `kb_audit/<ts>_netbox_push.yaml` row per write call.

Special Syntax in Interactive Mode¶

Syntax	Description
`@file.txt`	Inject file contents into the current prompt
`!command`	Execute a shell command directly (bypassing the Agent)

Comparison of the Three Interfaces¶

Interface	How to Start	Best For
CLI	`olav "query"`	Script integration, one-off queries, automation pipelines
TUI	`olav` (no arguments)	Multi-turn conversations, exploratory analysis, debugging
Web UI	`olav service web start` → `http://localhost:2280`	Browser access, team sharing