convez.ai
AgenteAutomaçãoIntermediário
vellum-ai

Vellum Assistant

vellum-ai/vellum-assistant

IA pessoal com memória e proatividade, roda local 24/7

Ver no GitHub
Muito bomPluginScore · de 0 a 100

Assistente de IA pessoal que funciona 24/7 gratuitamente em macOS, iOS, Web e integrado com Slack, Telegram e Twilio. Usa 8 tipos de memória para aprender seu fluxo de trabalho, atua proativamente entre aplicativos e pode ser autoexecutado. Agnóstico quanto ao modelo LLM utilizado.

Instalação

$npx convez-ai vellum-ai/vellum-assistant

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
1,5 mil
Forks
242
Acompanhando
9
Saúde
87%
Último commit
há 1 dia
Versão
v0.12.10
Issues abertas
129
Contribuidores
40

Casos de uso

  • Assistente pessoal que aprende suas preferências e padrões de trabalho
  • Automação de tarefas recorrentes com notificações proativas entre canais
  • Integração com Slack, Telegram, Twilio para um assistant onipresente

Veredito

Vellum é uma solução sólida para quem quer um assistente pessoal persistente com controle total. A arquitetura de memória híbrida e segurança fail-closed são destaques técnicos reais, mas exige familiaridade com autohosting e TypeScript para customização profunda.

README

A tradução para o português deste README ainda está na fila.


What it does

AreaSummary
MemoryLearns what matters and forgets what doesn't. Structured memory items — identity, preferences, projects, events — extracted with source attribution and deduplication. Hybrid retrieval (dense + sparse) ranks results semantically and lexically, with staleness windows per memory type. Per-user and per-channel isolation. Embeddings run locally by default.
IdentityBecomes its own. Behavior lives in SOUL.md, and during onboarding the assistant observes how you communicate and writes its own personality files. A per-user journal captures its reflections on past interactions. NOW.md acts as an ephemeral scratchpad for current focus and active threads.
ProactivityReaches out when something matters, without being asked. Every hour it checks in with itself: re-reads its notes, notices what's unfinished or due soon, and sends a message if needed. Notifications are routed to the right channel and won't interrupt you if you're already talking.
SecurityFail-closed by design. Actor identity is resolved once (guardian, trusted, or unknown) and enforced everywhere. Untrusted actors cannot read or write memory, trigger tools, or escalate. Credentials live in a separate process and never reach the model. Every tool runs in a sandbox.

Get started

1. Download the latest release

2. Open the app and pick your mode

  • Managed — sign in via Vellum Cloud, no local runtime required
  • Local — everything runs on your machine

3. Hatch your assistant

  • Give it a name, a personality, and the keys to your work

Prefer the terminal? See CLI install below.


Quick demo

https://github.com/user-attachments/assets/009bd0ae-95ac-4cf3-81bc-d54cd8631583


CLI

The CLI works but the desktop app is our primary focus. Available for advanced users, contributors, and non-macOS environments.

Install

bun install -g vellum
vellum hatch

Install from source

git clone https://github.com/vellum-ai/vellum-assistant.git
cd vellum-assistant
./setup.sh
source ~/.bashrc
vellum hatch

Common commands

vellum wake        # start services
vellum sleep       # stop services, keep data
vellum client      # interact through the terminal
vellum ps          # view running assistants
vellum terminal    # open a shell into a managed assistant container
vellum upgrade     # upgrade to latest version

All commands target the default assistant. If you have multiple, pass the assistant ID as the second argument.


Infra and security

AreaSummary
Trust engineDecides who can do what, and defaults to no. Fail-closed trust system that resolves actor identity once (guardian, trusted, or unknown) and enforces it everywhere. Untrusted actors cannot read or write memory, trigger tools, or escalate. Your credentials live in a separate process and never reach the model.
SkillsAdd new capabilities through sandboxed plugins. Manifest-driven plugins (SKILL.md + TOOLS.json) that inject tools and prompt sections at runtime. Skills can be bundled, installed from a catalog, or added from the workspace.
ChannelsOne assistant, everywhere you need it. Use it from the macOS app, Telegram, or Slack, with shared memory across all of them. More channels coming soon.
Multi-provider supportSwap models without changing anything else. Supports Anthropic Claude, OpenAI, Google Gemini, and Ollama for local models. Embeddings follow the same pattern: local ONNX by default, with automatic fallback to cloud providers.

Foundational documents

The canonical sources for who we are and how we talk about what we're building. The docs site at vellum.ai/docs is a rendered view of these files.

DocWhat it is
ConstitutionWho we are, what we believe, and what we refuse to compromise on
GlossaryThe shared vocabulary we use to talk about personal intelligence

Documentation

SectionWhat's covered
ArchitecturePlatform domains, repo structure, runtime · clients · gateway
Security & PermissionsSandbox, credentials, trust rules, permission modes
Features & CapabilitiesIntegrations, dynamic skills, browser, attachments, media embeds
API & CommunicationSSE event stream, event payloads, remote access
Development WorkflowClaude Code commands, parallel PRs, review loops, release pipeline

📖 Full documentation →


Contributing

We welcome contributions from everyone.

Community

License

MIT — see License. Integration logos from Simple Icons, licensed CC0 1.0.

Vellum Assistant is open-source software built by Vellum AI, a for-profit company. We also offer a managed product, the Vellum Platform, which sustains the business. Free to use and modify under MIT, and we're committed to keeping it that way.


Ver o original no GitHub

Bom para

Relacionados