Understanding knowledge

AI is becoming essential to our line of work increasingly capable, but also increasingly expensive. The goal of knowledge is to give Kalisio developers an AI assistant that is both context-aware and token-efficient, regardless of which coding agent they use.

Why not an existing solution

Several open-source projects already exist to index code and reduce token usage by improving context, such as claude-context. But none of them fit our needs:

• Paid dependency : solutions like claude-context require an OpenAI API key for embeddings, which goes against our open-source philosophy.
• Too agent-specific : most are built around a single coding agent. The field moves fast enough that today's dominant tool may not exist tomorrow; we don't want to lock ourselves in.
• Too generic : they treat all code the same. We want chunking and handlers tuned to our actual stack, .vue, .js, .json.

What knowledge provides

knowledge gives every coding agent the same structured context, through a ready-to-use configuration per agent and three complementary tools:

• Semantic code search : the Kalisio codebase is chunked and embedded into Qdrant.
• Git intelligence : commit history is parsed into hotspot scores, co-change patterns & bus-factor risk.
• Dependency graph : static AST analysis maps every import to know what depends on what.