Fundraising Fox

Eternis

San Francisco, US · 4 known investors

eternis.ai ↗

Eternis.ai is an AI research company building systems for forecasting, simulation, and calibrated prediction, including multi-agent decision-making tools for capital allocators, strategists, and forecasters. It also develops privacy-focused AI products that let users run models locally or in trusted execution environments so data stays under their control.

Also known as Eternis.ai · EternisAI

AI & Machine LearningDeep Tech

Investors · 4

Company profile

researched Aug 2026

Eternis (eternis.ai) describes itself as building frontier systems for forecasting, with the stated goal of forecasting, simulating and quantifying uncertainty using AI systems, and of understanding how knowledge and epistemics evolve. Its published research directions are: post-training language models for calibrated prediction, including pretraining on new modalities and maintaining cheap, continually learning world models; work on 'the shape of knowledge and metascience,' separating what a model knows from what it can fluently state, via epistemic embeddings, scaling laws and emergent properties; decision products that integrate this research into tools grounded in structured evidence; and multi-agent coordination, including multi-agent reinforcement learning, 'societies of twins' experiments and large-scale simulation.

The company operates a product ecosystem spanning four properties. Axion (axion.eternis.ai) is a multi-agent decision-making system aimed at capital allocators, strategists and others reasoning through high-stakes forecasts. Freysa (freysa.ai) is an agent that holds its own money and private keys and evolves over time, framed around governance layers for a human-AI future; a related strategy site is referenced at strategy.freysa.ai. Silo (siloprivacy.com) is a private AI application and hardware box that provides access to models hosted in trusted execution environments, with an option to purchase a box for local hosting. Lume (lume.top) is a prediction market platform for testing forecasts with real stakes.

Eternis maintains a public GitHub organization (EternisAI) with 37 public repositories, including 'enchanted-twin' (a Go digital-twin project, 92 stars), enchanted-proxy, an Axion forecasting skill portable across agent frameworks, Lume market-maker and SDK code, and confidential-computing components such as an attestation server, an attestation proxy for the Enclaver API, a fork of the Enclaver software-enclave toolkit, and a control-plane agent for confidential virtual machines.

Business model

Eternis positions itself as a research company whose research outputs feed a set of consumer- and professional-facing products: Axion (decision-making for capital allocators and strategists), Freysa (an autonomous agent holding its own funds and keys), Silo (a private AI app plus a purchasable hardware box for local or TEE-hosted model access) and Lume (a prediction market with real stakes). The sources do not disclose pricing, contracts or revenue arrangements beyond the statement that a Silo box can be purchased.

▸Full profile — history

History

Public material dates from 2026: the company site (copyright 2026) describes the research agenda and product ecosystem, and the blog's earliest listed post, 'Introducing Eternis,' is dated April 6, 2026, followed by research posts on March 24 and March 22, 2026. A GitHub organization, EternisAI, hosts 37 public repositories with commit activity through 2026, covering a digital twin application, TEE attestation tooling, confidential VM control-plane software, Lume market-making and SDK code, and an Axion forecasting skill. No funding, incorporation date or headcount is stated in the sources.

Compiled by commissioned research from 8 cited public sources — announcements, filings, and press listed under research sources below.

Key figures

latest reported
Forecasting model parameter countMar 20268,000,000,000 parameters
GitHub stars, enchanted-twin repositoryJan 202692 stars
Public GitHub repositories (EternisAI org)Jan 202637 repositories
Reported performance gain from hardest-example GRPO post-trainingMar 202647%

Company-reported or press-reported figures, each dated to when it was claimed — not independently audited.

Timeline · 5

launches, deals, and filings
Apr 2026
Blog post 'Introducing Eternis' outlines coordination-systems vision

Vision post stating the company builds coordination systems for a world in which humans, digital twins and autonomous agents participate in decisions, with forecasting as the foundation.

source ↗

Mar 2026
Research post: 'Towards SOTA Forecasting LLMs'

Eternis reports training an 8B-parameter model that it says surpasses all published baselines on open-ended forecasting, including models 10-15x larger, using context building, calibrated reward design and a modified GRPO procedure.

source ↗

Mar 2026
Research post: 'Hard Examples Are All You Need'

Post on maximizing GRPO post-training under annotation budgets, reporting up to 47% performance gains from focusing on the hardest examples.

source ↗

Jan 2026
Axion multi-agent decision-making system

Eternis lists Axion, a multi-agent system for decision-making aimed at capital allocators, strategists and others reasoning through high-stakes forecasts, at axion.eternis.ai. A related open-source 'axion-skill' repository describes a portable forecasting skill that returns research-backed probabilities for future events.

source ↗

Jan 2026
Product ecosystem: Freysa, Silo and Lume

Company site describes Freysa (an agent holding its own funds and private keys, freysa.ai), Silo (private AI app and hardware box with models hosted in TEEs, siloprivacy.com) and Lume (a prediction market platform, lume.top).

source ↗

Dated company events from announcements, filings, and press; legal rows summarize public dockets and regulator releases.

In the news

▸Research sources · 8

primary sources listed

8 public sources were cited for this profile; the first-party ones are listed here.

Frequently asked questions

What does Eternis do?
Eternis.ai is an AI research company building systems for forecasting, simulation, and calibrated prediction, including multi-agent decision-making tools for capital allocators, strategists, and forecasters. It also develops privacy-focused AI products that let users run models locally or in trusted execution environments so data stays under their control.
Who are Eternis's investors?
Eternis's investors include Arctic Digital, NIV, Pharsalus Capital, Coinbase Ventures.
Where is Eternis headquartered?
Eternis is headquartered in San Francisco, US.