Our purpose
The gap between a demo and an operator is infrastructure—and infrastructure is a research problem worth a whole lab.
Built by Rishit, in public, from India.
Principles for systems we can rely on.
Athrean is an independent AI product lab. Capability gets attention; accountable execution decides whether an agent is useful.
Standing rules
What we believe
does not move.
These constraints stay fixed even when models, tools, and implementation details change.
Evidence over confidence
A fluent completion claim is not proof. Real checks, rendered behavior, and a visible evidence chain decide whether work is done.
Boundaries before autonomy
Scope, budget, tools, and return contracts come before more agent freedom. Constraint is part of capability.
Systems over demos
The useful unit is not one impressive response. It is a run that stays oriented, handles failure, and finishes honestly.
Build in public
Working notes beat invented certainty. Product behavior, source, and open questions should remain inspectable.
Now and next
The work
on the table.
One product, several research directions, and a public record of what changes our mind.
Orchentra
Building the open-source, model-aware coding harness for constrained delegation and evidence-gated completion.
Context systems
Studying how long-running agents preserve trust-boundary instructions and fresh evidence through compaction.
Completion contracts
Turning a task's finish line into explicit checks that remain attached to execution and the final claim.
Publish working notes
Release research as claims, methods, and open questions that can be tested against implementation.
Built in public