AI teams produced $1.2B in audited output for Boulder startup Coheria in one year
Coheria’s collaborative AI platform assigns a different model family to check every piece of work and holds back
Press Release Disclaimer: This is a press release distributed through the XPR Media network. It has not been independently verified by our newsroom.
Coheria’s collaborative AI platform assigns a different model family to check every piece of work and holds back anything that skipped a required step
Fast facts
- What Coheria is: A collaborative AI platform that organizes open-source AI models into specialist teams that check each other’s work.
- How it checks work: Every team mixes model families, so the reviewing model never wrote the work. The Stickler feature enforces the best solutions through adversarial validation.
- Accountability: A real-time decision map records which model made which choice and why, in real time so you can jump in before it is too late.
- One-year output: 7.3 million distinct lines of code and 187,000 function points, by the company’s own count.
- Human role: Coheria’s two founders set architecture and direction. AI teams write, test and ship the code.
- Delivery: SaaS. Coheria owns, hosts and secures the platform.
- Planned products: Kindred for households, Inspire for small service businesses and Praxis for enterprises.
- Availability: Invite only Alpha access is planned for Q1 2027.
BOULDER, Colo., Oct. 15, 2026 (GLOBE NEWSWIRE) — Teams of open-source AI models produced $1.2B in audited results for startup Coheria in one year, according to an AI audit. Today, Coheria is introducing the collaborative AI team platform that did the work.
Each model on a Coheria team holds a specialist role, and the platform draws each team from different model families so the model reviewing a piece of work never wrote it. The models both collaborate and compete to produce the best answer and check each other’s output, a process Coheria co-founder and CEO Ken Tola calls adversarial validation. This process is controlled by a component called the Stickler, which holds back any work that skipped a required step, and that Stickler also provides a real-time decision map record of which model made which choice and why, allowing users to interact with the work before it is too late.
More than 23% of the total work completed over the past year was purely focused on these adversarial validation efforts, resulting in beyond-expert solutions. Tola and co-founder Keith Withington set architecture and direction, and the AI teams write, test and ship the code. Coheria fills human roles only where law or assurance requires a person.
Most AI products send each request to a single large model, which, in effect, checks its own work. Coheria gets its results from the way it organizes different types of reasoning models into beyond-expert teams, architected to ensure beyond-expert results.
“The fastest-growing line in most AI budgets is model API spend. Organizations buy access to frontier models priced to recover training runs you had no part in and will never see the books for. Coheria is built the other way around,” said Tola. “Capability comes from how the models are organized and work together rather than which model you can afford.”
Coheria plans three products on the platform: Kindred, a personal assistant for households; Inspire, a back office for small service businesses; and Praxis, AI teams for regulated enterprises, starting with banking.
Invite-only alpha access to Kindred and Inspire is expected in early 2027.
About Coheria
Coheria is a Boulder, Colo., artificial intelligence company. Its collaborative AI platform organizes open-source AI models into specialist teams that check each other’s work and record every decision. Ken Tola and Keith Withington founded the company in 2026. Its planned products are Kindred for households, Inspire for small service businesses and Praxis for regulated enterprises. More information is at coheria.ai.
Frequently asked questions
1. What is Coheria?
Coheria is a collaborative AI platform. It organizes open-source AI models into teams of specialists that compete to produce the best answer, check each other’s work and record every decision.
2. How is Coheria different from other AI products?
Most AI products rely on a single large model, which in effect reviews its own work. Coheria builds each team from different model families so the reviewer never wrote the work, and it depends on no single frontier model.
3. What is adversarial validation?
Tola’s term for how Coheria’s models attack each other’s output at every step. Mistakes, fabricated results and skipped steps get caught inside the team before anything reaches the customer.
4. What is a decision map?
A real-time trace of every decision the models make, showing the model, the step and the logic. When output is wrong, the map shows where and why and enables instant user engagement.
5. Who writes Coheria’s code?
The founders set architecture and direction, and AI teams exclusively write, test and ship the code.
6. Who owns and runs the platform?
Coheria owns, hosts and secures the platform and delivers it as SaaS. All data and output is owned by the end-user.
7. What products will run on it?
Kindred, a personal assistant for households; Inspire, a back office for small service businesses; and Praxis, AI teams for regulated enterprises, starting with banking. Alpha access for Kindred and Inspire opens in 2027.
Media Contact: doyle@prolexitypr.com (720) 523-3350
Media gallery
