Hire in one sentence
Say “I need a content lead for Xiaohongshu” and get an employee with a persona, skills, and boundaries. No prompt engineering, nothing to configure.
Alpha · open source · self-hostable
Hire a team of
AI employees.
Run them like a real company.
AgentPulse isn't another chatbot or a workflow builder. You hire AI employees, assign goals in a group chat, and let them discuss before they act — with you approving every high-risk move. If you can run a group chat, you can run a company.
- No prompts to write
- No workflow graphs
- Any OpenAI-compatible model
Most "AI tools" hand you a blank box and a blinking cursor. AgentPulse hands you a company. Departments, employees with names and memory, a task board, an inbox of decisions waiting for your call. The work doesn't vanish when you close the tab — it keeps running on a server while you sleep.
Watch a company think, not a bot reply.
You drop a goal in the group. The right employees get pulled in, argue out the unknowns, and turn the conversation into a consensus brief. Only after you confirm does work begin — and anything risky stops for your sign-off.
# next week — content plan
4 members
-
Say the goal
“Plan next week's content.” Plain language, in the group.
-
They discuss
Employees ask what's unclear instead of guessing — like real teammates.
-
You confirm the brief
The chat collapses into a consensus card. One click turns it into tasks.
-
They execute — you hold the keys
Publishing, spending, deploying: every high-risk action waits for your approval.
Built around how a company actually runs.
Discuss, then execute
Agents question fuzzy briefs in the group and reach a written consensus before touching anything.
The boss holds the keys
Publish, deploy, spend, send — every high-risk move stops and waits for allow-once / allow-always / deny.
Gets better the more you use it
Each employee keeps long-term memory and distills skills from its work — your corrections stick.
No idle employees
When there's no task, they reflect, spot opportunities, and drop ideas into an Idea Center — 24/7.
Everything leaves a trace
Who did what, how far along, what they produced — conversations scatter, tasks and outputs don't.
Not a chatbot. Not a workflow builder.
| Chatbot(ChatGPT-style) | Automation(Dify / n8n-style) | AgentPulse | |
|---|---|---|---|
| Mental model | One window, one assistant | Nodes, triggers, JSON | Run a company |
| Multi-role collaboration | ✕ single assistant | ~ you orchestrate | ✓ employees discuss & hand off |
| Process & output kept | ✕ gone when you close it | ~ engineer-facing | ✓ tasks, progress, outputs traced |
| Dangerous actions gated | ✕ | ~ you wire it | ✓ enforced approvals |
| Learning curve | Low | High | Low — hire & delegate in words |
For people who wear every hat.
Solo founders
One person covering marketing, ops, support and finance — hand it to an AI staff.
Creators
Topic research, multi-platform copy, performance reviews — run by an always-on operator.
Indie hackers
Outsource the repetitive business work to “digital employees”; stay on the core product.
Small teams
Add capacity without headcount, in the language of a company — not an API.
Open source, and yours to run.
The code is open for study and research. Each employee runs on the Hermes Agent runtime; any OpenAI-compatible model works (DeepSeek by default), and text-only models still handle images, audio and video.
Electron + React
FastAPI + PostgreSQL
Hermes runtime
DeepSeek / OpenAI-compatible
Read the source
Questions
How is this different from ChatGPT?
ChatGPT is one assistant in one window; close it and the context is gone. AgentPulse is a persistent team — each employee has its own persona, skills and memory, work is traced as tasks, and it runs 24/7 on a server without you keeping a window open.
Can the AI go rogue — publish or spend without me?
No. Every high-risk action (publishing, deploying, anything that spends money) is structurally stopped for your approval — allow once, allow always, or deny. Irreversible spending stays with you by design, not by asking the model nicely.
Do I need to be technical?
No. The product speaks in companies, employees, group chats, tasks and approvals — not prompts, workflows or DAGs. If you can run a group chat, you can run AgentPulse.
Which models does it support?
Any OpenAI-compatible API, DeepSeek by default. A text-only model still handles images, audio and video — the runtime converts every modality to text for the main model.
Is it ready to use?
It's Alpha. The desktop workspace, group chat, tasks, consensus briefs and the approval gate are working; real multi-agent execution is being wired in on the open-source runtime. Try it and tell us what breaks.
Start your one-person company —
with a full staff.
Clone it, run it locally, hire your first employee tonight.
git clone git@github.com:Clycheng/agentpulse.git