How we build, source, and fact-check the AI Echoes
Source-aware by design. Honest about what is documented and what is recreated.
In brief
Every figure in Agora Cosmica is an AI Echo. We distill a voice profile from the figure’s own primary works and their documented life, ground the conversation in that material, and ship a public factcheck for every figure that separates what is documented from what we recreated for the story. The Echo is an educational interpretation, openly labeled, never a recording and never the real person. The instruction profiles and the factchecks are public, and the whole codebase is open source, so you do not have to take our word for any of it.
How we build an Echo
We start with the person, not a persona. For each figure we read their primary works, the books, letters, and texts they actually wrote, alongside their documented life from standard biographies and scholarship. From that we distill a voice profile: how they reasoned, the ideas they returned to, the way they spoke. The voice profile is the recipe the model follows in a conversation. It is not a transcript of anything the person said.
What comes out is an AI Echo: an educational interpretation grounded in the source material. It is openly an Echo every time it speaks, never presented as a recording and never as the real person. No recordings of these figures talking to you exist. The synthesized audio is the figure’s Echo voice, a voice we give them, clearly separated from the original daguerreotypes, manuscripts, and writings that the work draws on.
A conversation is only one of the ways to learn here, meant for thinking an idea through. Most of what a figure offers is crafted ahead of time: a narrated life story, twelve core teachings, multi-voice dialogues. The live conversation sits alongside those, and it is built to send you onward to the primary texts, not to keep you talking.
Where the material comes from
The source of the words matters more than any single sentence. Because an Echo is AI-rendered, what it says will vary from one conversation to the next. The material underneath it does not. Four things make that material checkable.
Primary works
Each figure is anchored in their own writing, the texts they actually produced. The voice and ideas come from there first, before any modern summary.
A public factcheck
Every figure ships with a factcheck that lists what is historically documented and what was recreated for the narrative, shadow chapters included. It is published on the figure’s page, not hidden.
Open instruction profiles
The instruction profiles that ground each Echo are served from a public CDN, not locked inside the app. They are our signature on the method, and anyone can fetch and read them.
Open-source code
The whole platform is open source under AGPL-3.0. The way an Echo is assembled, sourced, and screened is in the repository for anyone to inspect.
How the fact-checking works
Every figure has a public factcheck, and it is specific. For each narrated chapter it names what is documented (dates, people, places, events drawn from the record) and what we recreated (a scene we staged, a line of dialogue, a sensory detail). Where the prose goes beyond the documented record, the factcheck says so plainly, and it labels the kind: an invented detail, a composite scene, a paraphrased source, a compressed timeline.
It does not flatter the figure. The factcheck includes a shadow section that records the harder parts of a life, the contradictions and the failures, because these were real and contradictory people. The aim is not a monument. It is an honest account you can check against the sources we list, which name the primary texts and the scholarship behind each figure.
This is the same obligation a careful historian takes on: reconstruct where you must, and disclose where you did. The factcheck is where we disclose it.
What an Echo can and cannot do
An Echo can be wrong. It is generated live, and no human reviews a sentence before it reaches you. It can misremember a date, blur a source, or state something with more confidence than the record supports. That is the honest limit of this medium, and we would rather name it than paper over it.
So treat an Echo as a learning companion, not an authority. It is a doorway into a body of work, useful for thinking out loud and finding where to look next. When something matters, verify it: open the figure’s factcheck, read the primary text it points to, and follow the sources. The best outcome is that you outgrow the Echo and go to the originals.
If you find something we got wrong, tell us at [email protected]. We will fix it, and the fix is public like everything else.
Why you can trust the method
You do not have to trust us. You can check. Four things hold the method accountable.
Nonprofit
Agora Cosmica is built by ChipMates gemeinnützige GmbH, a small German nonprofit. There are no shareholders and no engagement targets. A longer conversation earns us nothing and even costs us, so nothing here is tuned to keep you hooked.
Open source
The whole platform is open source under AGPL-3.0. The sourcing, the safety screening, and the privacy claims are all in the code, so the method is auditable rather than asserted.
No profiling
No tracking cookies, no third-party analytics, no profiles of who you are. Your chat history lives only in your browser, encrypted, and is never stored on our servers. Bring your own model key and it stays in your browser.
EU-hosted
Live speech runs on our own GPU servers in Germany, and the rest of the audio is served from EU hosting. You can also run the whole thing on your own machine.
Questions people ask
Is the AI the real person?
No. Every figure is an AI Echo, an educational interpretation grounded in that person’s primary works and documented life. It is clearly labeled as an Echo, never a recording and never the real person. No recordings of these figures speaking to you exist, and the synthesized audio is the figure’s Echo voice, not their own.
How accurate are the Echoes?
An Echo is grounded in the figure’s real work, so it stays close to the actual ideas. But it is generated live, with no human reviewing a sentence before it reaches you, so it can be wrong: a wrong date, a blurred source, more confidence than the record supports. Treat it as a learning companion, not an authority, and verify anything that matters against the sources.
Where do the facts come from?
Each figure is anchored in their own primary works, the books, letters, and texts they actually wrote, alongside standard biographies and scholarship. Every figure ships with a public factcheck that lists the sources and separates what is documented from what was recreated. The instruction profiles that ground each Echo are public, and the whole codebase is open source under AGPL-3.0.
What if it gets something wrong?
Open the figure’s factcheck, which names what is documented and what was recreated, and follow it to the primary text. If you find a genuine error, tell us at [email protected]. We fix it, and because the code and the factchecks are public, the correction is public too.
See it for yourself
Open the app, talk with an Echo, then open that figure’s factcheck and read the sources behind it. 30 free messages a day, no signup.
Start exploring If an honest method matters to you, a star is the simplest way to help us reach more people. Star on GitHub →Read further
This page covers how we build the Echoes. For why we give the dead a voice at all, the tradition behind it and where the risks lie, read why we call them Echoes. To see the method in a single life, browse the figures and open any factcheck.
To learn philosophy in conversation, start at the AI philosophy tutor. To learn timeless wisdom from history’s figures, start at the wisdom guide.
Or read the method in the code on GitHub.