Too many control panels
Every tool adds another surface to check. The work becomes fragmented across dashboards, status pages, prompts, and settings.
Project
ClawChat — Messaging for AI Agent Teams
Operate AI agents from a familiar chat interface.
ClawChat is a messaging layer for coordinating AI agents, workflows, files, approvals, and tool access. Instead of managing agents through dashboards and scattered settings panels, you talk to the right agent, give clear instructions, review what it has done, and decide what happens next.
Chat
User: Route last week's draft review to Content Flow.
Agent: Draft review queued. I'm running a style pass, checking facts, and preparing the approval notes.
Agent: Before I update the live document, which sources should I treat as authoritative?
Connected: Gmail, Research, Notion, Calendar
The problem
AI agents are becoming useful. The interface has not caught up.
Most agent products still feel like admin software: dashboards, configuration screens, disconnected logs, and hidden workflows. That may work for setup, but it is not how people naturally delegate work.
As soon as you have more than one agent, the real questions become harder: who is working on what, what context did they use, what action are they about to take, what needs human approval, and what happened last time?
Weak operating memory
Agents need more than chat history. They need durable instructions, project files, workflow rules, examples, and clear boundaries.
Supervision is too vague
If agents are going to touch real work, humans need visible activity, approval gates, and a simple way to intervene before anything important happens.
The ClawChat idea
ClawChat turns agent work into a managed conversation.
The core idea is simple: working with agents should feel closer to messaging a capable team than operating a software control room.
You describe the outcome. ClawChat helps route the work to the right agent or workflow. The agent uses its operating files, connected tools, and project context. You review the result, approve sensitive actions, and keep the work moving.
Message the agent, not the software
No special prompt builder. No buried workflow screen. You give instructions in the same place you see progress, questions, files, and outputs.
Give agents roles and operating files
Agents can be organised around responsibilities: research, content, software, operations, admin, finance, or project-specific workflows.
Keep humans in the loop
ClawChat is built around review and approval. Agents can prepare work, explain what they intend to do, and wait for permission before sensitive actions.
What ClawChat can connect to
A conversation layer above your working tools.
ClawChat is designed to sit above the systems where work already happens. The aim is not to replace every app. The aim is to give agents a controlled way to operate across them.
Designed for integrations with:
GmailCalendarGitHubLocal filesWeb researchInternal appsVideo and content toolsCustom business workflows
Use cases
What I'm building it for.
For the broader product context, see ClawChat as the public reference for the project.
Running AI agent teams
Coordinate specialist agents without losing track of responsibility, context, or status.
Operating business workflows
Give agents instructions, approval rules, and access to the systems needed to complete real tasks.
Content and research operations
Use agents to collect sources, summarise material, classify information, draft content, and prepare outputs.
Agent-operable software
Design apps so agents can actually use them, not just talk about them.
Product principles
Built around control, memory, and supervision.
- Messaging-first interaction
- Agent documentation as operating memory
- Human approval where it matters
- Visible logs of agent activity
- Clear permissions and workflow boundaries
- Local and private workflows where needed
- Built for real work, not demos
Visual showcase
Inside ClawChat
Chat
User: Route last week's draft review to Content Flow.
Agent: Draft review queued. I'm running a style pass, checking facts, and preparing the approval notes.
Agent: Before I update the live document, which sources should I treat as authoritative?
Connected: Gmail, Research, Notion, Calendar
Chat with agents
Give instructions, ask follow-up questions, and keep the working context in one conversation.
Chat
User: Route last week's draft review to Content Flow.
Agent: Draft review queued. I'm running a style pass, checking facts, and preparing the approval notes.
Agent: Before I update the live document, which sources should I treat as authoritative?
Connected: Gmail, Research, Notion, Calendar
Manage agent teams
Organise agents by role, project, department, or workflow so responsibility is clear.
Chat
User: Route last week's draft review to Content Flow.
Agent: Draft review queued. I'm running a style pass, checking facts, and preparing the approval notes.
Agent: Before I update the live document, which sources should I treat as authoritative?
Connected: Gmail, Research, Notion, Calendar
Connect apps and workflows
Give agents controlled access to the tools, files, and systems they need to do useful work.
Chat
User: Route last week's draft review to Content Flow.
Agent: Draft review queued. I'm running a style pass, checking facts, and preparing the approval notes.
Agent: Before I update the live document, which sources should I treat as authoritative?
Connected: Gmail, Research, Notion, Calendar
Track activity
See what agents are doing, what changed, and what still needs attention.
Chat
User: Route last week's draft review to Content Flow.
Agent: Draft review queued. I'm running a style pass, checking facts, and preparing the approval notes.
Agent: Before I update the live document, which sources should I treat as authoritative?
Connected: Gmail, Research, Notion, Calendar
Review before action
Approve sensitive steps before agents send, publish, update, delete, or commit anything important.
Chat
User: Route last week's draft review to Content Flow.
Agent: Draft review queued. I'm running a style pass, checking facts, and preparing the approval notes.
Agent: Before I update the live document, which sources should I treat as authoritative?
Connected: Gmail, Research, Notion, Calendar
Build agent operating packs
Turn instructions, examples, APIs, approval rules, and workflow docs into reusable operating memory.
Founder note
Why I'm building ClawChat
I'm building ClawChat because I do not think the future of AI agents is another dashboard.
If agents are going to become part of everyday work, they need an interface that feels natural to delegate through, but structured enough to manage properly. Messaging gives people the simple surface. Operating files, permissions, workflows, and approval gates give agents the structure.
ClawChat is my attempt to bring those two things together: a familiar conversation layer for serious agent work.
— Alex Kerss
Interested in ClawChat?
Interested in ClawChat?
I'm currently developing ClawChat and using it across my own AI, content, software, and business workflows.
If you're interested in the product direction, agent-operable software, or how messaging could become the interface for AI teams, feel free to get in touch. You can also explore Where apps become usable by AI agents.