Blog
Technical writing from the Sverklo team. Deep dives on hybrid code search, PageRank for source code, bi-temporal memory, and the engineering decisions behind the local-first MCP server.
Reciprocal Rank Fusion is doing 80% of the work in our hybrid search
I tried half a dozen scoring schemes for combining BM25, vector similarity, and PageRank. Most of them required tuning weights, calibrating scores, and explaining the result to skeptical reviewers. RRF is three lines of math, has no tunable parameters, and beats every alternative I tested. Here's what it does, why it works, and why it should be your default.
PageRank for source code: a 2026 revival
Embeddings tell you what code is similar; PageRank tells you what code is load-bearing. Neither alone is enough; together they're the difference between an LLM reading the test fixture and an LLM reading the production file. A short tour of an old idea applied to a new problem.
Bi-temporal memory for AI coding agents
When Claude Code compacts context, the decision you made about retry semantics last Tuesday is gone. Sverklo stores it against the git SHA it was made under and tells you whether the code it referred to still exists. This is why your agent should be git-aware.