This is where it started
“For multi-stage, branching AI processes, final output alone cannot explain the failing state, conflicting guardrail, or recurring transition problem.”observed → Started 2026.07.20
— So we built
AI Process Console
An observability console connecting AI process state and transitions through graphs, time series, and tests.
Developer evidence
What I owned in this project
- Period
- 2026.07 – Operating now
- Team
- Solo developer
- Contribution
- Planning · design · development · deployment · operations, 100%
- Owned scope
- Projects retain authority over execution and transition selection while the console observes past-tense facts only. Test requests are executed by an isolated TEST runner inside the project.
Introduction
AI Process Console reconstructs AI process state from project-emitted events and connects only authored transitions as node-edge-node graphs. It analyzes the same state evidence over time to locate guardrail conflicts and verifies changes through isolated state-based tests. PostHog and LangSmith metrics are collected for future unified site analysis.
User flow
- 01
Validate project-emitted fact events and reconstruct process state and authored transitions.
- 02
Connect state as node-edge-node in Overview and analyze time-series changes and guardrail conflicts.
- 03
Reproduce the selected state and transition in an isolated TEST run and compare a candidate with the baseline.
Highlights
- Visualizes event-observed process state through authored node-edge-node topology.
- Analyzes time-series state with guardrail verdicts to trace recurring conflicts.
- Runs isolated state-based tests and transition revalidation without mutating production.