About xeref.ai
Build your agents. Run your work. Let them talk to each other.
The Vision
Most AI tools give you a generic assistant. xeref.ai lets you build your own — agents that know your workflow, connect to your tools, remember your context, and coordinate with each other.
We believe the people who will get the most from AI are not those who use the most tools, but those who design systems tailored to how they actually work. xeref.ai gives you the methodology (CLAWS), the infrastructure (XerefHermes), the interface (Dashboard + MCP), and the deployment layer — so your agents are not isolated assistants but a coordinated team.
Built by Bugra Karsli — developer, content creator, and AI automation builder based in Turkey.
How Xeref Works
01 / Build
Agents without code
Pick from 48+ CLAWS capabilities in XerefClaw, generate a structured system prompt, and save it as a named agent. A working agent in minutes, not sprints.
Agents that coordinate
The XerefHermes message bus and scheduled workflows keep your agents talking — agents publish, handlers act, memory persists. Work keeps running when you are not watching.
Where your users already are
Ship agents to Telegram, Discord, WhatsApp, or a web chat widget — and reach the same workspace from Claude or Cursor via MCP. No new surface for your customers to learn.
The Platform
XerefClaw — Build
Browse 48+ agent capabilities organized by the CLAWS methodology (Connect, Listen, Archive, Wire, Sense, Agent Architecture). Select the capabilities you need, generate a structured system prompt, and save it as a named agent project. No code required.
Open XerefClaw →Dashboard — Manage
A full productivity environment: AI Chat with selectable system agents, Tasks with AI prioritization, Workflows (cron automations), Plans, Artifacts, Classroom, Memory, and a Hermes inspector for monitoring agent messages — all in one place.
Go to Dashboard →XerefHermes — Orchestrate
An inter-agent message bus built on a typed envelope system. Any agent or server-side routine can publish a message to the Hermes queue; registered handlers process it asynchronously. The built-in memory:save handler writes directly to your Pinecone long-term memory — more handlers ship with every release.
Deploy — Ship
Connect your agents to Telegram, Discord, WhatsApp, or a web chat widget. Set up cron routines that run your agent on a schedule and deliver output wherever you need it.
System Agents
XerefClaw
General AI assistant — CLAWS agent builder & productivity
Xeref Agents
Multi-agent architecture specialist — team design & orchestration
XerefHermes
Inter-agent message bus — publish & inspect Hermes envelopes
Selectable from the chat input — each agent brings a different system prompt and area of expertise.
The MCP Backbone
Every feature in the Dashboard — projects, tasks, notes, memory, daily targets — is also available as an MCP (Model Context Protocol) tool. Your Claude, Cursor, or Antigravity agent connects directly to your xeref workspace and works with your real data in real time.
What you manage in the UI is what your agent knows. There is no manual sync — the Dashboard, your agents, and the Hermes bus all share the same backend.
Read the MCP docs →Memory Architecture
Long-term memory is powered by Gemini Embedding 2 (3072-dimensional, 100+ languages) and Pinecone for vector storage. Documents, notes, and tasks you create are automatically embedded and semantically searchable — by you and by your agents. XerefHermes handlers can write to your personal memory namespace directly from agent workflows.
Personal memories are stored in isolated per-user namespaces. Your data is never mixed with other users' data.