Reflect is an open-source memory-graph-based self-reflection engine and backend. It helps a user build a persistent, transparent representation of their beliefs, observations, reflections, and conversations, and supports AI-assisted insight generation on top of that graph.
This repository is my continuation of work I originally began during the Data Portability Hackathon — sponsored by the AI Collective, AITX, the Data Transfer Initiative, and the University of Texas at Austin School of Law and Innovation — now consolidated under a unified name and released under an open-source license.
Going into the hackathon, I was already pursuing ideas in this domain through prior lines of work I called NTS (Number Theory of the Soul) and Journey. I was grateful for the opportunity to build something with collaborators under the high-pressure, time-constrained format of a competition, and the event gave me a reason to materialize those ideas into a working system quickly.
What I built was largely based on my intuitions rather than on formal studies of machine learning. Instead of designing a neural network from first principles to achieve the project's goals, I set out to build a harness around existing models — manipulating the system so that, given specified inputs from the user, it would produce outputs shaped the way I wanted, and then relationally storing both the user's inputs and the harness's outputs in a persistent knowledge graph.
That is, honestly, what this repository is: a working prototype of the harness-and-graph approach. In my own words, a bootleg version of what I'm ultimately trying to accomplish. I'm now devoting my studies to machine learning and attention mechanisms so that future iterations can replace the harness with a purpose-designed architecture that achieves the same goals natively, with fewer compromises.
I'm releasing this work openly in the spirit of the hackathon that hosted it — data portability, transparency, and the right to inspect and modify the systems that mediate one's own thinking. If it's useful to you, take it.
I built this with a collaborator, @clarkmyfancy. After the hackathon, my partner kept the ThriveSight name and the front-end they built, and has continued independent development under that brand. Everything in this repository is my portion of the work:
- the reasoning engine (signal generation, clustering, coordinate system, context assembly, graph agent, insight engine, LLM prompts, persona configuration)
- the Django REST backend that exposes it (authentication, workspace-scoped data, live conversation, graph persistence, analytics)
- the research corpus (memory hypotheses, signal address architecture, federated graph architecture, reflect graph contract family)
The ThriveSight name and the front-end belong to my partner and are not part of this repository.
reflect/
├── LICENSE Apache License 2.0
├── README.md this file
├── pyproject.toml Python package definition
├── thrivesight_core/ ← Python reasoning engine (internal name retained for import stability)
│ ├── signal_engine.py
│ ├── cluster_engine.py
│ ├── coordinate_system.py
│ ├── context_assembly.py
│ ├── graph_agent.py
│ ├── insight_engine.py
│ ├── llm_prompts.py
│ └── persona_config.py
├── api/ ← Django REST backend (squash-imported 2026-04-22 from tchjhnsn/reflect-api)
│ ├── manage.py
│ ├── requirements.txt
│ ├── reflect_api/ (Django project configuration + root URL wiring)
│ ├── events_api/ (HTTP transport, services, live-conversation, graph helpers, analytics)
│ ├── datasets/
│ ├── scripts/
│ └── …
└── _ideas/ idea notes (will migrate to docs/ideas/)
The internal Python package is still named thrivesight_core for import stability. Renaming it to reflect_core is tracked as an issue. The api/ subfolder was imported from the (now-archived) standalone tchjhnsn/reflect-api repo via a squash import — full per-commit history of the Django backend remains preserved at that archived repo.
Reflect is being consolidated into a single repository from previously separate codebases.
- ✅
api/— Django REST backend. Imported 2026-04-22 from the archived tchjhnsn/reflect-api. - ⏳
docs/— architecture, research, and thesis documents (currently in my vault; see the migration tracking issue). - ⏳
packages/— TypeScript domain packages (values ontology, scoring, number-theory framework) from the archived tchjhnsn/ai-lab.
Track the remaining consolidation progress in the issues.
pip install git+https://github.com/tchjhnsn/reflect.gitOr for local development:
git clone https://github.com/tchjhnsn/reflect.git
cd reflect
pip install -e .The engine requires Python 3.11+ and declares its dependencies (anthropic, neomodel, neo4j, jsonschema) via pyproject.toml.
- Python engine: released. Reference implementation; not actively deployed.
- Django backend: not yet merged; lives at tchjhnsn/reflect-api (archived).
- Docs corpus: not yet migrated into this repo.
I maintain this solo, at a best-effort cadence. Pull requests are welcome but may take time to review. Forks and derivative projects are explicitly encouraged — that's the point of releasing this.
- Hackathon context: the Data Portability Hackathon, sponsored by the AI Collective, AITX, the Data Transfer Initiative, and the University of Texas at Austin School of Law and Innovation. Thanks to the organizers for a format that made fast, honest building possible.
- Collaborator: @clarkmyfancy. Thank you for building alongside me. ThriveSight as a product and as a name is yours; the front-end you built continues its life under that brand, independent of this repository.
- Author: Torian Johnson (@tchjhnsn).
Apache License 2.0. In short: you can use, modify, and distribute this code for any purpose, including commercial. You must preserve the license notice and state any significant modifications. There is no warranty.