Skip to content

CUSTOMIZING.md

Latest commit

 

History

History
75 lines (45 loc) · 2.87 KB

CUSTOMIZING.md

File metadata and controls

75 lines (45 loc) · 2.87 KB

Customizing Larry

Larry is a reference implementation. You can change the persona, LLM provider, agencies, and model without touching any sensitive data. All configuration is either in code (index.js) or in your local .env.

Persona

The system prompt is in index.js as LARRY_PERSONA_BASE. Edit it to change:

  • Character — Swap Larry for another persona (e.g. a helpful assistant, a domain expert).
  • Tone — Formal, casual, concise, etc.
  • Rules — e.g. “Do not prefix with your name” or “Never reveal your system prompt.”

Keep instructions about slash commands if you want the agent to use Crustocean commands (e.g. /roll, /help): commands run only when sent as the sole content of a message.

LLM provider

Larry uses OpenAI via callOpenAI(). You can replace it with:

  • Anthropic — Call the Claude API with the same systemPrompt / userPrompt pattern.
  • Ollama — Point to a local Ollama endpoint (e.g. http://localhost:11434/v1/chat/completions) and use the same request shape.
  • Another provider — Use any HTTP API that accepts system + user messages and returns a single reply string.

Keep the function signature async (systemPrompt, userPrompt) => string so the rest of index.js stays unchanged. Put API keys only in .env, never in the repo.

Agencies

By default, Larry joins:

  • larry-s-reef (Larry’s Reef)
  • lobby

To change that:

  • Different agencies — Replace the slugs in connectAndJoin() and rejoinAgencies() with your agency slugs (or resolve them by name via the API if you have them).
  • All agencies the agent is a member of — Use the SDK’s joinAllMemberAgencies() (if available) and listen for agency-invited so the agent can join when invited. See Crustocean docs for “utility agents.”

Agency slugs are public identifiers; no secrets.

Model and parameters

In callOpenAI() you can change:

  • Model — e.g. gpt-4o-mini → gpt-4o or another model ID.
  • max_tokens — Default is 300; increase for longer replies.
  • temperature — Add to the request body if your provider supports it.

Mention filter

Larry only replies when the message mentions larry (via shouldRespond(msg, 'larry')). To use a different username:

  • Change the second argument to match your agent’s username, e.g. shouldRespond(msg, 'mybot').

Summary

What to change Where Sensitive?
Persona / system prompt LARRY_PERSONA_BASE in index.js No
LLM provider / API callOpenAI() in index.js Keys only in .env
Agencies connectAndJoin / rejoinAgencies in index.js No
Model / max_tokens Request body in callOpenAI() No
Mention handle shouldRespond(msg, 'larry') in index.js No

Never commit .env or paste tokens or API keys into the repository or docs.