Deep Dive

Technical exploration of every subsystem in the chati.dev orchestration engine. How context, memory, quality, and multi-provider coordination work under the hood.

System Overview

Full system architecture: pipeline phases, infrastructure layers, governance, and multi-provider support.

Agent Handoff Sequence: how agents communicate through the orchestrator with quality gate validation.

Explore Subsystems

PRISM Context Engine

LLMs lose context in long sessions

6-layer context assembly with progressive reinforcement. Budgets increase as memory degrades, scaled per provider.

Explore

Memory System

Knowledge dies with the session

3-level progressive retrieval persisting knowledge across sessions within the project, from volatile session state to long-term durable patterns.

Explore

Multi-Terminal & Multi-CLI

Locked into one AI provider

Parallel terminal execution, adapter pattern, overlay directories, and cost tracking across Claude, Gemini, and Codex.

Explore

Quality & Governance

AI code quality is inconsistent

5 pipeline gates, circuit breaker pattern, 3-tier verdicts, scoring thresholds, and safety net triggers.

Explore

Session Lifecycle

Workflows break between sessions

Session state machine, schema versioning, execution profiles, and mode transitions from planning to deployment.

Explore

The Constitution

AI agents operate without clear rules

19 governance articles with enforcement levels defining every rule of the orchestration system.

Explore

Design System

Design decisions scattered across files

Unified UX specification with design tokens, component patterns, icons, and tokenization audit.

Explore

By the Numbers

Specialized Agents

Constitution Articles

PRISM Layers

Quality Gates