Vectora

io.github.brunosrzPublished by GitHub account brunosrzv0.1.0rc3Updated Oct 24, 2026

AI assistant with RAG, web search, filesystem and memory. MCP sub-agent for Claude Code.

Overview

AI-generated overview

A self-hosted AI assistant MCP sub-agent that adds RAG, web search, filesystem access and memory to Claude Code.

What you can do
Vectora is a local MCP sub-agent for Claude Code that combines retrieval-augmented generation, web search and URL extraction, filesystem access, and memory. It is part of a self-hosted AI assistant app distributed as native installers rather than a container. The manifest does not list individual tools, so the exact tool surface is not documented here.
When to use it
Worth adding when you want an assistant to ground answers in your own documents and memory, search the web, and reach local files from within Claude Code. It suits users who are comfortable running a self-hosted Python-based assistant and supplying their own model and search API keys.
Requirements
Runs locally over stdio as the PyPI package vectora-agent, so a Python runtime is needed. Required secrets: COHERE_API_KEY for RAG embeddings and reranking, and TAVILY_API_KEY for web search and URL extraction. Optional secrets: GOOGLE_API_KEY (recommended free-tier LLM), ANTHROPIC_API_KEY, OPENAI_API_KEY, LANGSMITH_API_KEY. Desktop only; no web executable.
Before you install
It asks for several third-party API keys (COHERE_API_KEY, TAVILY_API_KEY, GOOGLE_API_KEY, ANTHROPIC_API_KEY, OPENAI_API_KEY, LANGSMITH_API_KEY); those keys are sent to their respective providers, and usage may incur costs. Filesystem access means the assistant can read or change local files, and web search and URL extraction send queries and URLs to Tavily. The project is proprietary and desktop only.

Installation

In SourceWeft

  1. Open Vectora in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

Vectora — Monorepo

Monorepo da Vectora. Cada produto vive em sua própria pasta, com toolchain e lockfile independentes. O que conecta tudo é o GitHub Actions (um workflow por projeto, gated por path) e o pre-commit compartilhado na raiz.

Estrutura

PastaProjetoStackDeploy
vectora/App Vectora (assistente de IA self-hosted)Python (uv) + Vite + ElectronInstaladores nativos (Nuitka + Electron) via canal de update
company/Site institucionalTanStack Start + Vite (pnpm)Vercel → vectora.company
docs/DocumentaçãoHugoVercel → docs.vectora.company
services/Gateway (OAuth/webhooks pro desktop, ex-relay) + updates (distribuição de releases) — era relay/ + update-server/, unificadosHono + Cloudflare Workers (pnpm)Cloudflare (wrangler)

O app em si (backend Python, frontend Vite, casca Electron) fica todo dentro de vectora/ — veja vectora/README.md para detalhes de arquitetura, build e CLI.

CI/CD (.github/workflows/)

4 workflows reais, cada um gated por path (on.push/pull_request.paths): uma mudança só no vectora/ não dispara o build/deploy de company/docs/services.

WorkflowDispara emFaz
vectora.ymlvectora/**, tags v*lint • security • build • testes (unit/integration/e2e) • release nativo (Nuitka + Electron) • publica no canal de update (R2 + KV)
edge.ymlcompany/**, services/**, docs/**lint/typecheck/test dos 3 projetos • deploy automático em push em master (Vercel + Cloudflare)
release-please.ymlpush em mastermantém o PR de release acumulado (Conventional Commits) — mesclar publica a tag vX.Y.Z
pr-checks.ymlpull requestslabels automáticos • valida título em Conventional Commits

Não existe workflow de Docker Build/Push, CodeQL nem triage de issues — a distribuição do app vectora é via instaladores nativos (não imagem Docker).

Secrets esperados

SecretUsado por
GOOGLE_API_KEY, COHERE_API_KEY, TAVILY_API_KEY, LLM_PROVIDER, LOG_LEVELtestes do vectora
WIN_CERTIFICATE_BASE64, WIN_CERTIFICATE_PASSWORD, APPLE_ID, APPLE_APP_SPECIFIC_PASSWORD, APPLE_TEAM_ID, VECTORA_RELEASES_TOKENassinatura/publicação dos instaladores nativos
VERCEL_TOKEN, VERCEL_ORG_ID, VERCEL_PROJECT_ID_COMPANY, VERCEL_PROJECT_ID_DOCSdeploy Vercel (company + docs)
CLOUDFLARE_API_TOKEN, CLOUDFLARE_ACCOUNT_ID, R2_ACCESS_KEY_ID, R2_SECRET_ACCESS_KEYdeploy do services + canal de update (R2)

Desenvolvimento

Cada projeto é independente — entre na pasta e use o gerenciador dele:

bash
# App Vectora (Python + Vite + Electron)cd vectora && uv sync && uv run vectora start
# Site institucionalcd company && pnpm install && pnpm dev
# Docs (Hugo)cd docs && hugo server
# Services (gateway + updates)cd services && pnpm install && pnpm dev

Pre-commit (raiz)

O .pre-commit-config.yaml na raiz cobre todos os projetos (ruff/ty/bandit no vectora/, prettier/oxlint/tsc no frontend, actionlint nos workflows):

bash
pre-commit installpre-commit run --all-files

Licença

Proprietária. Consulte LICENSE.

Source: README.md at commit bcdc09d

Tools

0
SourceWeft has not started this package in its sandbox yet, so its tools are not known.

Version history

1
  1. v0.1.0rc3LatestSep 16, 2026

Score

40% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 days25 pts

    Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.

  • Auth declared10 pts

    Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.

  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts

    Declare a license in the repository or the registry entry.