ClaraVerse

## What is ClaraVerse? ClaraVerse is a private AI workspace built around three things most AI chat UIs treat as an afterthought: agents that work as an actual team, memory that behaves like a real memory system instead of a flat list of notes, and a license that does not start charging you once people notice your project. Chat, Crew (multi-agent teams with human review), a visual workflow builder, and Telegram integration all live in one app, on your own infrastructure. Use OpenAI, Claude, Gemini, or local models like Ollama and llama.cpp. Chats are stored locally on your device by default, with optional encrypted sync if you want them on more than one machine. If you have **Ollama** or **LM Studio** running on your machine, ClaraVerse detects them automatically and imports all your models. Zero configuration. --- ## Quick Start ### One-line install (Linux / macOS) ```bash curl -fsSL https://raw.githubusercontent.com/claraverse-space/ClaraVerse/main/cli/install.sh | bash && claraverse init ``` ### Docker Compose ```bash git clone https://github.com/claraverse-space/ClaraVerse.git cd ClaraVerse docker compose -f docker-compose.production.yml up -d ``` Open **http://localhost:3000**, register your account (first user becomes admin), and start chatting. > **Have Ollama running?** ClaraVerse auto-detects it and imports all your models. No setup needed. > Make sure Ollama is listening on `0.0.0.0` (set `OLLAMA_HOST=0.0.0.0` in your Ollama config). ### Want it in your terminal too? ```bash claraverse agent install # builds Clara Agent, puts `claracli` on your PATH claracli # then run /login claraverse to connect it ``` See [Clara Agent](#clara-agent-the-same-assistant-in-your-terminal) below. Docker run (single command) ```bash docker run -d \ --name claraverse \ -p 3000:3000 \ -v claraverse-data:/app/data \ -v claraverse-uploads:/app/uploads \ --add-host=host.docker.internal:host-gateway \ ghcr.io/claraverse-space/claraverse:latest ``` This runs ClaraVerse with a single container. For the full stack with MySQL, MongoDB, Redis, SearXNG, **Qdrant and the embeddings sidecar (required for Knowledge bases / RAG)**, use the Docker Compose setup above. Single-container mode boots fine, but the Knowledge tab and `search_knowledge` tool need the sidecars and will surface "embeddings service unreachable" without them. Advanced configuration ### Prerequisites - Docker & Docker Compose v2+ - 4 GB RAM minimum (8 GB recommended) - Ollama, LM Studio, or any OpenAI-compatible API ### Environment variables Override defaults by creating a `.env` file next to `docker-compose.production.yml`: ```bash # Port (default 3000) CLARAVERSE_PORT=8080 # Point to a custom Ollama or LM Studio URL OLLAMA_BASE_URL=http://host.docker.internal:11434 LMSTUDIO_BASE_URL=http://host.docker.internal:1234 ``` ### Local AI auto-detection ClaraVerse automatically discovers local AI providers running on your host machine: | Provider | Default URL | What happens | | ------------------- | ------------------------------------- | ------------------------------------------------- | | **Ollama** | `http://host.docker.internal:11434` | Models imported, provider created, visibility set | | **LM Studio** | `http://host.docker.internal:1234` | Models imported via OpenAI-compatible API | The discovery runs every 2 minutes. When a provider goes offline, it's automatically disabled. When it comes back, models are re-imported. **Ollama setup tip:** Ollama defaults to `127.0.0.1`, which Docker containers can't reach. Set `OLLAMA_HOST=0.0.0.0`: ```bash # If using systemd: sudo systemctl edit ollama # Add under [Service]: # Environment="OLLAMA_HOST=0.0.0.0" sudo systemctl restart ollama ``` ### Troubleshooting ```bash # View logs docker compose -f docker-compose.production.yml logs -f claraverse # Restart docker compose -f docker-compose.production.yml restart # Fresh start (removes all data) docker compose -f docker-compose.production.yml down -v && docker compose -f docker-compose.production.yml up -d ``` --- ## Core Capabilities ### Crew: Agent Teams With Human Review Give a project a brief, hire a team of agents, and work the card pipeline. Every card comes back for your review before it ships, so you always know what's being worked on and why. No black box. ### Memory: Layered, Encrypted, and Actually Used Clara remembers you across conversations without being told twice. Facts are extracted automatically in the background, pulled back in only when relevant to the current conversation, and every entry is visible and editable in Settings. Nothing is stored that you can't see or delete. - **Pinned tier**: mark a fact as always-inject (allergies, hard constraints, how you want to be addressed). Pinned memories skip relevance scoring and never decay. - **Recall tier**: everything else, retrieved by embedding similarity against the current conversation so only what's relevant gets injected. - **Decay**: unused memories lose relevance over time and archive themselves automatically. - **Model-driven**: Clara calls `search_memory` and `add_memory` mid-conversation, the same way it calls any other tool. - **Encrypted at rest**: AES-256-GCM with a key derived per user via HKDF. Not even a ClaraVerse admin can read your memories.

### Clara Agent: The Same Assistant, In Your Terminal [Clara Agent](clara-agent/) is a coding agent that runs in your terminal and ships in this repo. It reads, writes, and edits files and runs commands on your machine, using the same model and account as the web app. ```bash claraverse agent install # builds it and puts `claracli` on your PATH claracli # start it in any project directory ``` Then run `/login claraverse` inside the agent. You get a short code to confirm in the browser, and it picks up whichever model your account is configured with. The machine then shows up under **Settings → Devices**, where you can revoke it at any time. It talks directly to your configured model provider, so it keeps working even when the ClaraVerse server is down, and it makes no network calls to anything except your own instance and your own model provider. ### Skills in Chat Clara uses skills mid-conversation: context-aware tools that activate when needed. Search the web, generate images, analyze data, all without leaving the chat.

### Channels: Telegram Integration Talk to Clara from Telegram when you're away from the app. Set up routines that run on a schedule and report back to your phone.

### 150+ Integrations & Tools Out of the Box Slack, GitHub, Jira, Google Sheets, Notion, Discord, Telegram, HubSpot, and many more, built in, no MCP required. All integrations are shared across Chat, Workflows, Crew, and Routines. Connect once, use everywhere.

### Interactive Artifacts Have all your creations in one place: images, charts, games, apps, and more.

### AI Docs Built-in tools for PPT, PDF, CSV, and much more.

### Interactive Chat UI (Human in the Loop) The AI asks you visual questions when it needs your input.

### Workflows: Visual Automation Drag-and-drop workflow builder with parallel execution, scheduling, and 200+ integrations. Describe what you need and let the LLM build the automation for you.

### Schedule, API and Automate Use workflows with your web apps, schedule daily messages, automate repetitive tasks.

--- ## ClaraVerse vs Open WebUI Open WebUI is the biggest self-hosted AI chat UI, and a good one. It has a bigger community, more vector database integrations, and a deeper document extraction pipeline than we do. We're not pretending otherwise. But it is a chat UI with tools bolted on. ClaraVerse is built differently, and it shows in three places. | | ClaraVerse | Open WebUI | | --- | --- | --- | | **License** | AGPL-3.0. Host it, white-label it, sell access to it. The only condition is that your modifications to the code stay open source. | BSD-3 with a branding clause added in v0.6.6. Once you pass 50 users in any 30-day window, you're required to keep "Open WebUI" branding visible unless you buy an enterprise license. | | **Multi-agent teams** | Crew ships in the box: hire a team of agents, work moves through a Kanban pipeline (Drafts, Queued, Working, Review, Done), and every card comes back to a human before it ships. | No equivalent. "Agents" are per-model system prompt and tool presets. Kanban-style multi-agent orchestration isn't part of the product. | | **Memory** | Layered by default. A pinned tier that never decays and always gets injected, a recall tier retrieved by embedding similarity, automatic decay and archival for stale entries, AES-256-GCM encryption per user. The model calls it as a tool, same as any other tool. | Manual notes, or an optional background job that extracts facts every few turns. The tiering and relevance logic you'd actually want exists as a third-party community Function you install yourself. No documented encryption model. | | **Workflow automation** | Visual drag-and-drop builder with parallel branches and scheduling. Describe what you want and the builder generates it for you. | Scheduled prompt runs and structured task lists. No visual node-based builder. | | **Local model support** | Auto-detects Ollama, LM Studio, and any OpenAI-compatible endpoint, including llama.cpp, on a 2-minute poll. | Auto-detects Ollama. | | **Terminal agent** | Clara Agent ships in the repo. One command installs `claracli`, which shares your account, model, and device list with the web app. | No first-party terminal coding agent. | If you need enterprise SSO, SCIM provisioning, and a large plugin marketplace today, Open WebUI is more mature there. If you want agents that act like a team, memory that behaves like an actual memory system, and a license that doesn't ask you to pay once you succeed, that's what ClaraVerse is for. --- ## Key Features | Feature | Description | | --------------------------------- | ---------------------------------------------------------------------------- | | **Crew** | Agent teams with a card pipeline and human review | | **Clara Agent** | Terminal coding agent (`claracli`) sharing your account and model, ships in-repo | | **Memory System** | Layered pinned/recall memory, AES-256-GCM encrypted, model-driven retrieval, automatic decay | | **Knowledge bases** | Upload PDFs / MD / TXT / HTML to a project, search via hybrid vector + BM25 + reranker. Available in Chat and Workflows | | **Skills** | Context-aware tools that activate mid-conversation when needed | | **Channels** | Telegram integration, talk to Clara from your phone | | **Routines** | Scheduled task sequences that report back via Telegram | | **Workflows** | Drag-and-drop builder with parallel execution, scheduling, 200+ integrations | | **150+ Integrations** | Slack, GitHub, Jira, Notion, and more, shared across Chat, Workflows, Crew, and Routines | | **Devices** | Connect all your machines, Clara reaches MCP on any of them remotely | | **Local-First Storage** | Conversations live in IndexedDB by default, with optional encrypted cloud sync | | **Local AI Auto-Detection** | Ollama and LM Studio discovered and imported automatically |