Architecture
A control plane between AI and organisational state.
People and agents interact with ReasonTrail between external AI providers and the organisational record. Context, authorization, execution, verification, and memory updates cross explicit boundaries.
System overview
Data model
A project-centered relationship map.
Organizations define the tenant boundary. Projects connect captured evidence, structured work, context versions, reviews, and controlled Operator events.
Provider-independent governance
Change the model without losing control.
Claude, OpenAI, Gemini, DeepSeek, local models, and custom agents can change while the governed organisational record remains. Router AI keeps provider credentials server-side.
context.version: versionedtoken_budget: boundedprovider: configureddatabase_credentials: never sentOperator lifecycle
One controlled agent, persisted stages.
Operator v1 supports controlled task-status work and bounded repository inspection. It does not expose arbitrary shell, deployment, browser, or external-system actions.
Multi-agent continuity
Handoff through reviewed project memory.
One agent leaves project knowledge and evidence; ReasonTrail stores and reviews it; the next agent retrieves bounded context and continues. Coordinated multi-worker execution remains a future milestone.
Read the current architecture