AI agents can collect evidence, compare sources, draft findings and coordinate parts of a research workflow. That makes work faster, but it also creates a governance problem: if many agents contribute to an output, an organisation still needs to know which claims are supported, which records are controlled and what is approved for publication.
AgentBrain approaches this as an intelligence-governance problem. The system is not only designed to produce outputs. It is designed to preserve the difference between work performed, evidence reviewed, records accepted and material released publicly.
What AgentBrain is.
AgentBrain is a controlled intelligence system for persistent AI work. In the AMR research environment, it coordinates memory, source-backed evidence, confidence boundaries, review steps and explicit publication approval.
The purpose is not to make every agent output authoritative. The purpose is to make useful work inspectable before it becomes part of a public or decision-grade record.
The process.
Governance principles.
Work is separated before synthesis.
Specialist workers can operate in parallel, but their outputs do not automatically become final findings.
Unsupported claims can be stopped.
A promising conclusion cannot override the evidence boundary that supports it.
Opportunity does not overrule evidence.
Commercial or narrative usefulness is not treated as a substitute for source support.
Internal records remain auditable.
Accepted records keep their status, scope and review context visible for later work.
External publication is separate.
Public-facing outputs require explicit approval rather than automatic release.
Novelty claims stay conservative.
The system avoids first-ever, unique, unprecedented and world-first wording unless independently supportable.
What this release represents.
The current AMR public evidence release is a traceability environment. It shows how a frozen research base, edge-level methodology and claim review can be presented for institutional inspection.
It should be read as a structured evidence review, not as a real-time surveillance product, an automated AMR monitor or a causal inference engine. Traceability, implementation quality and effectiveness are separate analytical questions.
Whether a public evidence chain can be reconstructed.
Whether capital, activity or evidence can be defensibly assigned to an AMR claim.
Whether a measurable result can be connected to a specific intervention.
Evidence rules.
Traceability is not performance.
A visible evidence chain does not prove implementation quality or effectiveness.
Not assessed is not broken.
Insufficient public evidence is not the same as proof of failure.
Missing public evidence is not missing activity.
Non-public work may exist even when the public chain cannot verify it.
Hard claims need locators.
Every strong claim needs evidence, source location, denominator and status.
About this release.
Published through the Human Publication Gate. Outputs remain auditable at amr.agentbrain.ch.
- Dashboard
- v0.2.2
- Research base
- ALPHA-MEGA-001 v1.0.1, frozen
- Methodology
- Frozen edge-level traceability with claim-level reproducibility
- Contact
- research@agentbrain.ch
Review the AMR evidence environment.
Explore the public evidence chain, country pages and methodology behind the current release.
Explore Evidence