← all repositories
claraverse-space/ClaraVerse

Self-hosted AI workspace that actually talks to your tools

ClaraVerse bundles chat, workflows, task queues, and 150+ integrations into one privacy-focused platform you can run locally.

ClaraVerse
Velocity · 7d
+8.4
★ / day
Trend
steady
star history

What it does

ClaraVerse is a self-hosted AI workspace — think chat, visual workflow builder, long-running task management (“Nexus”), and Telegram integration all in one Go/React app. It plugs into OpenAI, Anthropic, Google, or local models via Ollama/LM Studio, and stores conversations in browser-local IndexedDB so the server never sees your chat history.

The interesting bit

The auto-detection of local AI is genuinely frictionless: if Ollama or LM Studio is running, ClaraVerse discovers it every 2 minutes, imports models, disables the provider when it goes offline, and re-imports when it returns. No manual provider configuration. The “Clara Companion” CLI also bridges local MCP servers to the instance over WebSocket, so your local filesystem and tools become remote-accessible skills.

Key highlights

  • Nexus task queue: Assign research/coding tasks, track on a Kanban board — the AI reports progress rather than vanishing into a black box
  • 150+ native integrations: Slack, GitHub, Jira, Notion, etc., shared across Chat/Workflows/Nexus/Routines without requiring MCP
  • Knowledge bases: Hybrid search (vector + BM25 + reranker) over uploaded PDFs/MD/TXT, usable in chat and workflows
  • Browser-local storage: Conversations stay in IndexedDB; server operates zero-knowledge for chat data
  • Telegram bridge: Talk to Clara from your phone, get scheduled routine reports back
  • AGPL-3.0 licensed, with Docker Compose bundling MySQL, MongoDB, Redis, SearXNG, and Qdrant

Caveats

  • Single-container Docker mode works for basic chat, but Knowledge bases and the search_knowledge tool require the full Docker Compose stack with Qdrant and an embeddings sidecar — otherwise you get “embeddings service unreachable”
  • 4 GB RAM minimum, 8 GB recommended; this is not a lightweight Raspberry Pi project
  • Ollama defaults to 127.0.0.1 which Docker can’t reach — you must rebind to 0.0.0.0 for auto-detection to function

Verdict

Worth a look if you want a unified AI workspace under your own infrastructure and are willing to pay the Docker complexity tax. Skip it if you just need a chat frontend for Ollama — this is substantially more (and heavier) than that.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.