ntxt.ai
ntxt provides shared memory for AI assistants like Claude, Cursor, and ChatGPT through a context graph that captures decisions, failed experiments, and dynamic project state. The platform connects via a single URL and token to any MCP-compatible tool, allowing AI assistants to read from and write to a structured knowledge graph. Original research shows the graph retrieves latest committed state more reliably than chat search, which mixes in superseded versions.
- Developer Tools
- Artificial Intelligence
- Software Only
Funding
Founders
Founder details are not available yet.
Product
Problem
AI assistants have no persistent memory between sessions, forcing users to re-explain project context, decisions, and failed experiments every time. Static context files like CLAUDE.md capture static preferences but miss the dynamic state of a project, the rationale for decisions, and the history of what was tried and abandoned. This re-priming wastes minutes per session across every tool and still fails to give AI a complete picture of the project's trajectory.
Solution
ntxt provides shared memory for AI assistants like Claude, Cursor, and ChatGPT through a context graph that captures decisions, failed experiments, and dynamic project state. The platform connects via a single URL and token to any MCP-compatible tool, allowing AI assistants to read from and write to a structured knowledge graph. Unlike static files, ntxt stores the "what" and "why" of project decisions, with a planned dedicated rationale field to force capturing reasoning alongside outcomes. The context graph is designed to surface the latest committed state, superseding older decisions, so AI assistants retrieve current and accurate information about pricing, strategy, and technical choices. This persistent memory enables AI tools to remember what was tried, why it failed, and what the current state is, without requiring users to re-explain context each session.
Target Audience
Primary users are developers, product builders, and power users of AI coding assistants and chat tools who need persistent project context across sessions and tools.
Features
- Context graph with structured nodes capturing decisions, pricing changes, messaging pivots, and product state, with versioning that marks superseded decisions
- MCP protocol integration via a single URL and token for reading and writing to the graph from Claude, Cursor, and ChatGPT
- Original research benchmarking shows the graph retrieves latest committed state more reliably than chat search, which mixes in superseded versions
- Planned dedicated rationale field to capture the reasoning behind decisions alongside outcomes
- Supports dynamic project state
- subreddit scraping lists, homepage copy, onboarding design
- that static files like CLAUDE.md cannot capture