Mixus blends inbox simplicity with enterprise-grade AI agents to put human-in-the-loop automation within reach — no code or custom frameworks required
A New Take on AI Agents: Start With the Inbox
AI agents are billed as the future of work—but most are clunky, opaque, or only work in sandboxes. Mixus, a startup launched out of Stanford in late 2024, believes it has a more realistic approach: AI agents that run inside tools you already use, starting with email and Slack.
“We’re meeting customers where they are today,” said co-founder Elliot Katz. “Where is every person in the workforce today? Email.”
With $2.6 million in pre-seed funding, and early customers like Rainbow Shops, Mixus is positioning itself as the “collaborative agent” layer that integrates with existing workflows, not tries to replace them.
From “Build Your Own Agent” to “Email It a Prompt”
Unlike agent frameworks like LangChain, AutoGen, or CrewAI, which require technical setup, Mixus enables users to spin up agents via plain-language email.
- Just send a prompt to
[email protected], and the system will build, run, and schedule an agent based on your instructions. - Example: Ask Mixus to review Jira tasks, draft emails to overdue assignees, request your approval, and repeat every Monday.
“We’re trying to democratize access,” said Katz. “You don’t need a developer background to use Mixus.”
Human-in-the-Loop by Design
AI agents that act autonomously are great in theory—but in practice, humans still need oversight. That’s where Mixus shines.
- You can set checkpoints where the agent pauses for your review.
- Add teammates to the workflow simply by CC’ing them or tagging them in the agent chat.
- Organizations can even set policy-level rules (e.g., humans must review outbound emails).
This makes Mixus agents safer, more transparent, and more collaborative than many on the market today.
Shared Memory and Multi-Agent Context
A standout feature of Mixus is Spaces—shared memory hubs that allow teams to collaborate with agents using persistent, collective knowledge.
- Files, prompts, conversations, and users are contextually linked.
- Agents can access this memory to reference prior tasks or documents without needing re-instruction.
- Unlike ChatGPT and Claude, Mixus offers cross-user memory sharing, even on early versions.
This enables agents that can truly work across teams, not just with a single user.
Agents That Act More Like Teammates
Mixus agents can do more than read emails or edit documents—they can:
- Integrate with tools like Jira, Gmail, and Google Sheets
- Run on a schedule or be triggered in real time
- Autonomously research, summarize, or even determine project ownership inside organizational systems
- Perform inline document editing, similar to Google Workspace or Copilot tools—but without being sandboxed
Agents are powered by OpenAI’s o3 model and Anthropic’s Claude 4, with access to the live web for research or monitoring.
“It’s like Google Alerts on steroids,” said co-founder Shai Magzimof.
If It Works, It Could Reshape Workflow Automation
Mixus is still in beta, but its demo shows capable, memory-rich agents that feel more like collaborative coworkers than passive assistants.
- That could make Mixus one of the first platforms to deliver usable AI agents at scale, without the infrastructure overhead or technical complexity of existing systems.
- Its inbox-first approach may be just the bridge that AI agents need to become truly mainstream.
