Context for coding agents
Glaux shows hidden coupling, surfaces files an agent would miss, and explains what to verify before a change lands.
The Problem
A coding agent reads the file it was asked to change. It misses the files that have always changed alongside it. It doesn't know this file is a coordination point for three other modules. It makes the edit. The PR surprises the reviewer.
01
Edits one file
The agent reads the file it was asked to change and nothing else.
02
Misses hidden dependencies
Three other files always change with this one. No import connects them. The agent doesn't know.
03
Causes follow-up fixes
The change lands. Review catches the missing updates. The cycle repeats.
The Solution
One MCP call returns a decision-ready briefing: risk verdict, hidden coupling, files to read first, and verification hints. The agent gets full context before touching anything.
01
Select a file
Agent picks the file it's about to change.
02
Get risk verdict
Glaux returns a risk level and confidence score instantly.
03
See missing files
Hidden coupling, expected partners, dormant dependencies — all surfaced.
04
Verify the right areas
Deterministic check types: cross-module regression, downstream callers.
Why It Works
Glaux combines git history, structural dependencies, and semantic understanding into a single context layer. No single source of truth reveals what all three together do.
Git graph
What changes together
Co-change patterns across thousands of commits reveal behavioral coupling that no import graph captures.
Static graph
What depends on what
Tree-sitter static analysis resolves imports, symbol calls, and inheritance across the whole repo.
Semantic graph
What the code means
Community detection clusters files into logical modules. PageRank surfaces which files are structural hubs.
Glaux
Turns all three into one decision-ready context block for the agent.
What It Produces
Glaux doesn't return centrality scores and pagerank. It returns decisions.
Files You Should Review First
most actionableHidden coupling: files that co-change with no import path. Expected partners. Dormant dependencies.
Why This File Is Risky
explains the riskPrimary drivers with evidence: hidden coupling, top 4% centrality, 94% paired change rate.
What To Verify
actionable checksDeterministic check types: cross-module regression, downstream callers, paired-file review.
Change Pattern
behavioral contextHow often this file changes, whether it ever changes alone, its paired change rate.
What The Agent Sees
full transparencyThe exact context block Glaux delivers via MCP: changeRisk, reviewFirst, primaryDrivers, suggestedCheckTypes.
Risk Verdict
decision-firstHigh / medium / low with confidence score. One line tells the agent how carefully to proceed.
Built for coding agents
Any agent that speaks MCP gets full change context automatically. No prompt engineering. No manual documentation. One tool call.
Also available as a REST API for custom integrations, dashboards, and CI pipelines.
Early Access
Want to see Glaux running on your repo? Reach out and we'll get you set up.