Codebase intelligence, explained
Guides on auto-generated documentation, MCP tools for AI agents, git intelligence, and how repowise compares to other tools.
All posts
Same refactor, 27× fewer tokens with Repowise MCP
See how Repowise MCP cut the same refactor to 27× fewer tokens, with transcript and tool-call traces showing the exact savings.
Leiden communities vs PageRank for code hotspots
Use leiden community detection to separate hotspots from boundaries in a code graph. Compare it with PageRank and pick the right signal for one repo decision.
Claude Code vs Cursor on a Python import cycle
Claude Code vs Cursor on one Python import cycle refactor: see the exact prompts, where each agent stalled, and which handled the graph better.
Codebase RAG buys you less than git history and ownership
Why the codebase RAG landscape matters less than git history and ownership for a 50-person team. See the 4 questions that cut false leads.
Security review on a Python monolith with 19 hotspots
A security review codebase case study on a Python monolith: find 19 hotspots, map ownership gaps, and cut duplicate reads with one review path.
Generate claude.md from a real repo, not a blank page
Generate claude.md from repo architecture, ownership, and recent changes with MCP calls. Give Claude Code a day-one brief that cuts guessing.
RAG for code is not embeddings plus a vector store
RAG for code fails when it stops at chunks and vectors. See a 4-layer model with symbols, history, ownership, and a 36% cheaper retrieval path.
Claude Code vs Cursor on a flaky Flask refactor
Claude Code vs Cursor on one Flask refactor: same prompt, exact edits, and where each agent lost the thread. See the repo context that changed the outcome.
When an MCP tool should return raw text, not answers
A practical mcp tools design rule: return raw evidence when the model needs to inspect files, symbols, or decisions. See the boundary in Repowise.
Why we ship a graph-aware MCP server instead of file readers
Graph aware mcp cuts grep loops: resolve callers, dependents, and review impact in one hop. See the token and latency math.
Task-Shaped MCP Tools Beat Entity CRUD for Code Agents
See mcp tools design on a real regression: task-shaped tools cut re-reading and context loss, with 49% fewer calls and 27× fewer tokens.
The dead-code detector we shipped with Pagerank and git history
Code hotspot detection as ranking, not a linter: Pagerank plus git history sorted quiet files from risky ones, so you can ignore the right 3.