AGEMS — Agent Management System
An operating system for AI-native businesses — build, deploy and orchestrate teams of autonomous agents under human oversight.
How this started
AGEMS is an open-source platform for running teams of autonomous AI agents — an “operating system for AI-native businesses,” where agents with different roles collaborate under human approval. We co-developed it with SeeGuru.
Challenge
Single AI calls are easy; a team of agents that plans, delegates, uses tools and stays under human control is not. It needs real infrastructure: multi-provider models, live communication, tool access, approval policies, and an audit trail you can trust.
Approach
We built AGEMS to treat agents like an org: roles, an org chart, role-based access, tiered human-in-the-loop approvals and audit trails — with real-time WebSocket communication between agents and humans, pluggable LLM providers (OpenAI, Anthropic, Google, Mistral, Ollama…) and tool hooks (APIs, databases, MCP, S3). Open-source and self-hostable via Docker.
The hard decision
We made it open-source and self-hostable rather than a closed SaaS. It's a slower path to revenue, but for an “operating system” businesses run their agents on, trust and control matter more than lock-in — you can read the code and host it yourself.
Agents modelled as an organisation
A single AI call is easy. A team of agents that plans, delegates, uses tools and stays under human control is not — it needs real infrastructure. AGEMS treats agents the way you'd treat a company: roles, an org chart, role-based access, and tiered human-in-the-loop approvals, all on an auditable trail.
Underneath, it's provider-agnostic and tool-agnostic:
- pluggable LLM providers — OpenAI, Anthropic, Google, Mistral, Ollama and more;
- real-time communication between agents and the humans supervising them;
- tool hooks into APIs, databases, MCP and S3;
- tiered approvals, so a human signs off on the actions that matter.
Open by design
It's open-source and self-hostable via Docker. For an "operating system" businesses run their agents on, we'd rather earn trust with code you can read and host yourself than lock it behind a black box.
“An operating system for AI-native businesses — where agents collaborate like an org, under human sign-off.”
— AGEMS, co-developed with SeeGuru
Before / After
Before
Ad-hoc scripts
After
Org with roles
Multiple agents
Before
After the fact
After
Tiered approvals
Human oversight
Before
Locked in
After
Pluggable
Provider choice
The Impact
Quantitative
Qualitative
Agents modelled as an org — roles, access, approvals, audit.
Multi-LLM and tool-agnostic (APIs, DBs, MCP, S3).
Open, inspectable and self-hostable — not a black box.
We can actually see and approve what the agents are doing. That's the difference between a toy and something you'd run a business on.
SeeGuru (joint development) · Founder, AI-native operations
