CortexGraph is a markdown-native personal context graph and agent context graph where your notes, sources, decisions, and agents share the same durable context. It gives humans and AI systems a common memory layer they can actually build on together.
Even when agents record history, it's siloed in their specific platform. You ask a model for help, get an answer, and the context fades into chat history or a proprietary database. Prompts get longer. Cross-collaboration becomes impossible.
Our architecture is completely harness-agnostic. Let ChatGPT read a thread, have a self-hosted Ollama model summarize it, and let a custom autonomous agent execute the result—all sharing the exact same reality.
Built on plain markdown files. Portable, inspectable, and future-proof.
Preserve who did what, what changed, and what decisions were made.
Use your vault across tools and clients through the Model Context Protocol. Connect external agents to a shared context graph.
Ingest notes, web research, and external sources.
Turn raw material into typed primitives and shared threads.
Let multiple agents update notes and leave checkpoints.
Return to active threads and unresolved work with full context.
When autonomous workers hit a clarity gap, they shouldn't guess or hallucinate. Instead, they ping the context graph. Cortex then proactively reaches out to you via iMessage to clarify ambiguities, feeding the answer back to the agent.