Local repo graph
Analyze repository structure locally and derive a project graph without sending the source repository to the server.
AI context infrastructure
ContextHub
One living context layer for every coding agent.
Find what your agents do not know. Answer it once. Keep product, architecture, requirements and plans current across tools and teams.
Local repo analysisConfidence-gated questionsWeb + CLIAgent-native context
The problem
AI coding agents are capable. Their project understanding is fragile.
Teams maintain CLAUDE.md, AGENTS.md, tool rules, architecture notes and requirements independently. The same fact can drift across files, disappear from a conversation, or remain an assumption nobody has challenged.
File conversion is not the hard problem. Freshness and missing context are. ContextHub builds a canonical project understanding, finds low-confidence gaps, and asks the team the questions that matter.
Product surface
Built around the loop.
ContextHub treats context as a living system instead of another document.
Ranked context questions
Low-confidence facts become explicit questions ranked by impact and confidence gap.
One canonical context
Product, architecture, requirements, constraints and answers strengthen one shared context layer.
Plan against the graph
Planning uses the same project context and unresolved questions rather than a disconnected chat history.
Agent-native rendering
Synchronize the canonical context into the files and shapes different coding agents expect.
Exportable knowledge
The accumulated context layer remains portable instead of becoming a black-box dependency.
How it works
From repository to current agent context.
01
Analyze locally
Build the project graph on the developer's machine. Source code stays local.
02
Expose uncertainty
Turn incomplete assumptions into ranked, answerable questions.
03
Answer anywhere
A human can resolve the question from the web or CLI. The answer updates the canonical context.
04
Synchronize agents
Render the current understanding into each agent's native context surface.
05
Close the loop
Plans and completed work flow back into the same project understanding.
context / checkout-serviceconfidence loop
Q1What happens after the third payment retry?importance 0.94 · confidence 0.31ANSWERMove to manual review; never re-authorize automatically.canonical context updatedSYNC4 native agent contexts regenerated locallyfreshContextHub
One living context layer for every coding agent.
Questions about access, product collaboration or the underlying system?