@personize/cli

Official Personize CLI. Manage guidelines, memory, agents, MCPs, and collections from the terminal or from AI agents that can shell out.

Install

npm install -g @personize/cli
personize setup   # interactive auth + org selection

The CLI persists your API key + selected organization in your user config. Subsequent commands pick them up automatically.

Auth

# Set an API key non-interactively
personize auth login --api-key sk_live_...

# Inspect the current session
personize auth whoami

The CLI also honors PERSONIZE_API_KEY and PERSONIZE_API_URL environment variables when set.

prompt

Run an AI prompt with the same execution model as the SDK's client.ai.prompt. Supports governed-memory mode (parallel auto-retrieve + deterministic system prompt), native web search, opt-in MCPs, multi-step instructions, structured output extraction, server-side evaluation, and auto-memorize.

Governed-memory + native-search flags (0.10.x)

| Flag | Purpose | |---|---| | --governed-memory / --no-governed-memory | Toggle governed-memory orchestration (parallel memory.retrieve + context.retrieve + deterministic system prompt). Default is org/SSM-driven. | | --agent-tools / --no-agent-tools | Toggle the executor's full agent tool registry. Default is !governed-memory. | | --auto-recall / --no-auto-recall | Run memory.retrieve in the parallel-retrieve step. Honored only with --governed-memory. Default on. | | --auto-guidelines / --no-auto-guidelines | Run context.retrieve in the parallel-retrieve step. Honored only with --governed-memory. Default on. | | --web-search / --no-web-search | Enable native provider web search inside the LLM step. Default on for BYOK, off for platform-billed. | | --web-search-max-uses <n> | Max native-search calls per step. | | --web-search-allowed-domains <list> | Comma-separated allowlist of domains for web search. | | --web-search-blocked-domains <list> | Comma-separated blocklist of domains for web search. | | --mcps <list> | Comma-separated MCP integration ids to load (allowlist). | | --all-mcps | Load all org MCPs (mcps: true). | | --no-mcps | Load no MCPs (mcps: false). |

Other prompt flags

--instruction <text>, --instructions-file <path>, --context <text>, --context-file <path>, --tier <basic|pro|ultra> (default pro), --model, --provider, --outputs <names>, --evaluate, --evaluate-criteria <criteria>, --memorize-email, --memorize-website, --memorize-record-id, --memorize-type, --capture-tool-results, --session-id, --attachment <path> (repeatable), --stream.

Examples

# 1) Cheap default — governed-memory, no agent tools.
personize prompt \
  --instruction 'Draft a follow-up to [email protected] about the renewal' \
  --governed-memory --no-agent-tools

# 2) Governed-memory with native web search and a citation-bounded allowlist.
personize prompt \
  --instruction "Summarize this week's announcements from Anthropic" \
  --governed-memory --no-agent-tools \
  --web-search --web-search-max-uses 3 \
  --web-search-allowed-domains anthropic.com

# 3) Governed-memory with an opt-in MCP integration loaded for tool use.
personize prompt \
  --instruction 'Find recent HubSpot contacts in the Acme deal and draft outreach' \
  --governed-memory --agent-tools \
  --mcps hubspot

Streaming works with the same flags by adding --stream. Use personize prompt --help for the full, current flag list.

Other command groups

The CLI exposes the same surface area as the SDK. Group help is available via personize <group> --help:

agents, agentdocs, analytics, auth, chat, collections, context, destinations, doctor, entities, evaluate, filter, guidelines, mcp, mcps, members, memory, notifications, organizations, prompt, responses, schema, setup.

v1.1 command group (additive)

The personize v1.1 * namespace wraps the SDK's client.v1_1.* methods (the additive v1.1 API surface that sunsets v1 on 2026-07-15). Existing v1 commands are unchanged.

| Command | Wraps | |---|---| | personize v1.1 memory save | POST /api/v1.1/memory/save (unified shape: shortform | document) | | personize v1.1 memory save-batch | POST /api/v1.1/memory/save/batch (per-item results) | | personize v1.1 memory retrieve [query] | POST /api/v1.1/memory/retrieve (modes: scout | brief | expand | filter | fetch) | | personize v1.1 memory manage list | GET /api/v1.1/memory/manage | | personize v1.1 context doc-types list | GET /api/v1.1/context/manage/doc-types |

Run personize v1.1 --help for the latest list. Remaining methods (manage CRUD, keys CRUD, import, doc-types CRUD, context save-batch) are TODO follow-ups — the SDK methods exist; CLI wrappers will be added in subsequent releases.

Output formats

Most commands accept --format json|table|yaml and --fields <names> to project specific columns. --quiet suppresses progress chatter for scripting.

Links

• Docs: https://personize.ai/technical-guide
• SDK: https://www.npmjs.com/package/@personize/sdk
• Issues: https://github.com/personizeai/personize-cli/issues