How xeref.ai works

What is xeref.ai?

xeref.ai is an AI agent builder and productivity platform. It has two layers: XerefClaw — a visual builder for designing and generating custom AI agents — and the Dashboard, a productivity environment where you manage tasks, chat with AI, run workflows, and connect your agents to real channels like Telegram.

How the layers connect

XerefClaw generates the system prompt that defines your agent's personality and capabilities. That prompt is saved as a project and used by the Dashboard's AI Chat as the agent's system context. Everything you build in XerefClaw powers the agents you run in the Dashboard.

1. Open XerefClaw

Click "XerefClaw" on the landing page or navigate to /builder. No account is required to browse and configure features.

2. Select Your Features

Browse 48+ capabilities organized across the CLAWS categories. Click any card to add it to your basket. Use the search bar or category filter to narrow down options.

3. Generate Your Prompt

Once you've selected your features, open the basket and click "Generate Prompt". The output is a structured system prompt ready to paste into Antigravity IDE or save to your xeref.ai project.

4. Save Your Configuration

Sign in with Google or a magic link to save named agent projects. Saved projects appear in your Dashboard under Agents and can be used as the context for AI Chat sessions.

Connect

Integrations that let your agent communicate: Telegram, Discord, Slack, WhatsApp, email, and more. Start here to define your agent's input/output channels.

Listen

Perception capabilities: voice transcription, document and image parsing, OCR, video analysis, speaker diarization, and language translation. Controls what raw information your agent can ingest beyond plain text.

Archive

Memory and persistence: vector stores (Pinecone, Qdrant), conversation history, knowledge bases. Gives your agent long-term context.

Wire

External tool integrations: Notion, GitHub, Google Calendar, Google Sheets, Stripe, Linear/Jira, search APIs, HTTP/REST tools, and outbound webhooks. Defines what your agent can act on.

Sense

Proactive behaviors and scheduling: cron jobs, inbound webhook triggers, RSS feed monitoring, inbox watching, and change detection. Turns a reactive agent into one that initiates on its own schedule.

Agent Architecture

Core agent identity and reasoning: personality definition, multi-agent routing, reflection loops, guardrails, and structured output. Configures the agent as an identity before any channel or tool is added.

Home

Your command center: saved agents, daily targets, and a snapshot of active tasks. Set up to 3 daily goals each morning — they surface here and inform AI task prioritization throughout the day.

Tasks

A unified task list with AI-powered prioritization. Type a goal and xeref generates a full project plan with phases and tasks. Ask "what should I work on next?" to get a reasoned top-3 based on priorities, deadlines, and daily targets.

AI Chat

Chat with any of your saved agents. The conversation uses the agent's generated system prompt as context, so each chat session is specialized to a specific agent you built in XerefClaw. You can paste a URL into chat and the assistant will fetch and read the page for you. On Pro and Ultra plans, attach an image or paste an image URL and the assistant will analyse it using a vision model. Model access depends on your plan (see Plans & Model Access).

Memory

Upload documents (PDF, text, images) and xeref chunks and embeds them into your personal Pinecone vector store. During AI Chat, relevant memory chunks are retrieved and injected as context automatically. You can also add manual memory entries and search across everything you've stored.

Artifacts

A versioned store for outputs your agents produce: code snippets, documents, prompts, data files, images, and workflows. Each artifact tracks its full version history and can be published with a share URL.

Plans

Generate structured execution plans from a single goal. xeref breaks it into phases, tasks, roles, and KPIs — giving you a ready-to-execute project blueprint you can track in the Tasks view.

Calendar

View deadlines, focus blocks, and scheduled events. Syncs with Google Calendar. Tasks with due dates appear on the calendar automatically.

Workflows

Cron-triggered automations. Define a schedule and a prompt — xeref runs the agent on your schedule and delivers output to your configured channel (Telegram, email, etc.).

Stats

Productivity trends, task completion rates, and AI usage metrics over time.

Agents

All saved agent projects from XerefClaw. Start a chat from any agent, deploy it to a channel, or edit its configuration.

Connectors

Manage OAuth connections and local integrations at /customize/connectors. The connector panel is organized into three groups: Web (GitHub, Gmail, Google Calendar, Notion, Slack, Supabase, Custom Webhook, Vercel), Channels (Telegram Bot), and Desktop (Filesystem). Connected services show a blue dot. Click any connector to view its detail panel, manage scopes, or disconnect.

Web connectors

GitHub, Gmail, Google Calendar, Notion, Slack, and Vercel use OAuth — click Connect and authorize in the provider's window. The connection token is stored encrypted (AES-256-GCM) server-side. Supabase and Custom Webhook are configured manually via token/URL entry.

Telegram Bot

Connect a Telegram bot by pasting your bot token in Dashboard → Deploy. Once connected, messages sent to your bot are routed to your configured xeref agent and replies are sent back through Telegram. The connector shows your masked token and a Disconnect action.

Filesystem (Desktop)

Give your agents local file access via the built-in MCP Filesystem server. Add allowed directories by path, then configure per-tool permissions (Always allow / Ask each time / Never allow) for read, write, and delete operations. Settings persist in your browser's local storage.

Skills

Browse and manage agent skills at /customize/skills. Skills are reusable instruction sets that extend what your agents can do — search for a skill, view its source, and add it to a project. Skills are stored in your profile and can be attached to any agent project.

Sidebar customization

Control which views appear in your dashboard sidebar via the "Customize Sidebar" button at the bottom of the nav. Toggle visibility, reorder items with ↑/↓, and hide items you don't use — they collect under a "More" group. Preferences are saved to your profile and persist across devices.

Getting your API key

Go to Dashboard → Settings → API. Your personal API key and MCP endpoint URL are shown there. Copy both — the endpoint is what you paste into your editor config, the key goes in the Authorization header.

Connecting VS Code, Cursor, or Windsurf

Add the Xeref MCP server to your editor's MCP settings. In Cursor and Windsurf, open Settings → MCP Servers and paste the config below. In VS Code with an MCP extension, add it to your mcp_config.json.

{
  "mcpServers": {
    "xeref": {
      "url": "https://xeref.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer <your-api-key>"
      }
    }
  }
}

Connecting Claude Code

Run the following in your terminal to register Xeref as an MCP server in Claude Code. Replace <your-api-key> with the key from Dashboard → Settings → API.

claude mcp add xeref \
  --transport http \
  --url https://xeref.ai/api/mcp \
  --header "Authorization: Bearer <your-api-key>"

Connecting Antigravity

In Antigravity IDE, open the MCP panel and click "Add server". Select HTTP transport, enter https://xeref.ai/api/mcp as the URL, and add the Authorization header with your API key. Your Xeref workspace is immediately available as context.

Connecting Codex CLI

Add Xeref to your Codex MCP configuration file (typically ~/.codex/config.json). Use the same endpoint and Authorization header pattern shown above.

What is the MCP backend?

Every piece of data in xeref.ai — projects, tasks, memories, daily targets — is exposed as an MCP (Model Context Protocol) tool. Your Claude, Antigravity, Codex, or any MCP-compatible agent can connect to this server and read, write, and reason over your workspace directly.

Finding your endpoint

Your personal MCP endpoint and API key are in Dashboard → Settings → API. The endpoint is unique to your account — treat the API key like a password.

Available tools

list_projects, create_task, update_task, list_memories, save_memory, recall_memories, delete_memory, set_daily_targets, suggest_next_task, create_project_with_plan, rename_project, list_routines, and more. Full tool reference is in the API tab of your Dashboard settings.

Basic (Free)

Access to XerefClaw, Dashboard home, tasks, and AI Chat with the xeref base model. No credit card required.

Pro — $17/mo or $170/yr

Everything in Basic, plus Claude Haiku 4.5, Claude Sonnet 4.6, and DeepSeek V4 Flash model access in AI Chat, higher workflow run limits, and priority support.

Ultra — $77/mo or $770/yr

Everything in Pro, plus all available models (including DeepSeek V4 Pro), unlimited workflow runs, and full memory/embedding storage.

Upgrading

Go to Dashboard → Settings → Billing, or visit the Pricing page. Upgrades take effect immediately via Creem.

Check the FAQ

Visit the FAQ page for answers to common questions about the platform, data privacy, and feature compatibility.

Contact support

Email support@xeref.ai — we typically respond within 24 hours.