---
name: vellum-self-knowledge
description: Answer questions about Vellum, the assistant's architecture, and its current configuration including which model and provider are active
compatibility: "Designed for Vellum personal assistants"
metadata:
  emoji: "🪞"
  vellum:
    display-name: "Vellum Self-Knowledge"
    activation-hints:
      - "When the user asks what model the assistant is running on"
      - "When the user asks about Vellum, how the assistant works, or its architecture"
      - "When the user asks about the assistant's current configuration or settings"
    avoid-when:
      - "When the user wants to change configuration (use in-chat config instead)"
---

## What is Vellum

Vellum is a personal AI assistant platform that runs as a local service on the user's machine. It supports multiple LLM providers (Anthropic, OpenAI, Gemini, Ollama, Fireworks, OpenRouter) and is accessible via a macOS desktop app, voice calls, and messaging channels like Telegram.

## Architecture at a Glance

The assistant runs as an HTTP server. Conversations are managed by a coordinator that delegates to an AgentLoop, which sends messages to the configured LLM provider and executes tools. The system prompt is composed from workspace files (IDENTITY.md, SOUL.md, and the guardian's per-user persona at `users/<slug>.md`) plus dynamic context. Skills extend capabilities via lazy-loaded instruction sets.

## Configuration System

Config is stored in `config.json` in the workspace directory. Use the CLI to interact with it:

- **Read a value**: `assistant config get <key>`
- **Set a value**: `assistant config set <key> <value>`
- **Search config**: `assistant config list --search <query>`

## When to Consult References

Consult these reference files for detailed knowledge on specific topics. Read the relevant file before answering questions in that domain.

- `references/inference.md` — Model identity, provider details, model catalog, and how inference routing works. **Read this when asked what model you are or about inference configuration.**

## Current Assistant Info

!`bun run "{baseDir}/scripts/self-info.ts"`

## Critical Rule

Never guess or hallucinate information about yourself. If unsure, consult the relevant reference file. The current assistant info above is populated at skill-load time and reflects the live configuration.