Doppelganger LabsBook a call

The platform: Doppelganger

One harness. Twelve modules.

Not twelve products — one idea, split twelve ways on purpose. Decisions, work, knowledge, permissions, tools and learning each get their own program, their own storage and their own owner. So your judgment lands where it matters, and the system can get better without quietly granting itself more authority. Every licence includes the control core — Rules & Approvals and the Operator Inbox. The other ten are yours to add when the work demands them.

Control core — in every licence

Rules & Approvals

The safety switch. Everything that matters asks it first; no answer means no action. It can suggest a safer rule — it can never grant itself one.

Operator Inbox

Your decision queue. Approvals, escalations, and sign-offs stack here — with the reasoning attached — while work continues. Where it's heading: a fleet console — many agents, many environments, one queue.

Work & knowledge

Work

One board for you and your agents — projects, goals, tasks, comments, and the evidence behind finished work. You all see who's on what.

Knowledge

The memory: documents, sources, research and history, with where every answer came from still attached. Next: a notebook research experience your agents learn into.

Observer

Watches the sources you connect, turns the noise into signals, and escalates what needs you. It notices — it never decides. More sources come online with every release.

Workflows

Something you do repeatedly becomes a reusable routine: recorded, evidenced, versioned, and published only once you approve it. The goal: it spots the pattern and proposes the routine itself.

Reach & tools

Marketplace

1,000+ connections to your tools, with permissions as narrow as you want — allow "send message" without allowing "delete channel". How deep each connection goes varies by provider.

Local Runtime Bridge

The hands, and their limits. It reaches the files and local capabilities you name — access to a file is never access to the machine.

Remote Setup

Plans the installation on your server: every service, connection, and safety check reviewed before anything runs. Planning and running stay separate on purpose.

Creation & collaboration

Open Design

Design work made by you and your agents together — files, conversations, runs and outputs, with the reasoning behind them kept.

Live Artifacts

What your agents produce lands here as something usable and governed — a small tool or preview carrying its origin and current state, not a file in a folder. The aim: design to artifact to connected tool, one unbroken flow.

Mesh

A reviewed map of the agents, services and peers around you, and what each one is trusted with. Discovery is not endorsement — nothing is trusted because it was found. Where it's heading: trusted collaboration across machines, teams and organizations.

Quick answers.

What is an AI agent operations platform?

Software for running AI agents you actually rely on. Doppelganger gives agents bounded capabilities instead of raw credentials, holds consequential actions for human approval, and keeps every decision attributable and recoverable — on servers you control, under rules you set.

Why twelve modules instead of one assistant?

Because you shouldn't have to hand all of your work, knowledge, access and automation to one opaque system. Decisions, work, knowledge, permissions, tools and learning each get their own program, their own storage and their own owner. That's why a person lands at the decisions that matter, why the system can get better without quietly widening its own permissions, and why you can start with part of it. The build direction is a fleet: many agents, shared identity, one decision queue, the same fail-closed gate.

Do I need all twelve modules?

No — and you shouldn't start there. Every licence includes the control core (Rules & Approvals and the Operator Inbox); most people add Knowledge and Work next. Each module is its own program, so what you didn't buy simply isn't running.

Can the AI engine be swapped?

Yes. The harness is the permanent layer; the engine underneath is your choice and can change without rebuilding anything you've set up.

What does self-hosted mean here?

Every module, database, and memory runs on a server you control — your own hardware or a cloud account that's yours — behind a private network. Nothing lives in our cloud.

How Doppelganger compares to ChatGPT →

The modules chain: Observer notices, Work plans it, Knowledge supplies the context, Marketplace supplies the tools, Rules & Approvals checks what's allowed — and the Operator Inbox asks you only when judgment is needed. Install it yourself from source, or have a forward-deployed engineer do it — see how a deployment works →

Last updated: September 5, 2026