FEATURED
Context that travels_
Cursor's indexer is excellent inside Cursor. Maguyva is the hosted MCP map the same agent can query from other editors, CLIs, and CI.
[Ecosystem]
[MCP]
[Portability]
> ls /var/log/decisions/*.md
Technical notes from the Maguyva build
Architecture decisions, MCP implementation notes, and the occasional rant about things that should not be this hard.
Latest dispatch
Most recent build note
Start here if you want the freshest technical update before you drop into the archive.
Archive
Technical notes by topic
Architecture, observability, MCP surface design, and the patterns that survived contact with production.
Map first, review second_
Ground the coding agent in a repo map before it writes. Let Greptile review the pull request after. A complementary pairing, not a swap.
[Ecosystem]
[Workflow]
[Code Review]
Give an AI agent context of a large repo without a local indexer_
How to give a coding agent a queryable map of a large repository from the cloud, without running a local indexer on your laptop.
[AI Agents]
[MCP]
[Indexing]
[Monorepos]
Why your AI coding agent hallucinates file paths_
AI coding agents invent file paths when they have no real map of the repo. Here is why it happens, and how grounding leads to fewer hallucinated edits.
[AI Agents]
[Hallucination]
[Retrieval]
[MCP]
Why We Upgraded Code Search to voyage-4-large_
We moved our code embeddings to voyage-4-large — currently top of the public RTEB code retrieval leaderboard. The honest version: the trade we make, what we actually index, and why we pay for premium embeddings.
[Embeddings]
[Search]
[Architecture]
Language Recursive Self-Improvement: Grinding Code Intelligence Across ~280 Languages_
We support code intelligence for ~280 languages. No human can hand-audit that. So we built a language recursive self-improvement loop — spot-check, LLM-as-judge, fix one thing, re-validate — and run it with a fleet of isolated agents until extraction is actually right, not just green.
[Architecture]
[Languages]
[Agents]
Multi-Modal Fusion Search: Picking the Right Retriever For Every Query_
A query like 'where is parseConfig defined' wants a different search than 'how does auth work'. Maguyva classifies the intent, weights four retrieval modalities accordingly, and fuses the results with weighted Reciprocal Rank Fusion.
[Search]
[Architecture]
Agent Observability: Hooks, Alloy, and Grafana_
We wired Claude Code and Codex into one Grafana stack with OpenTelemetry and Alloy, then used traces and logs to find and fix agent behavior issues at the source.
[Observability]
[Grafana]
[OpenTelemetry]
[Architecture]
Mining the Loop: How Changes Become Institutional Memory_
Git commits become structured changelog entries and architectural decision records, then feed back into AI agents as queryable institutional memory.
[Architecture]
[Workflows]
Skill Mining: From 3,500 Candidates to 466 Capabilities_
We screened 3,500 skill candidates and adopted 466. A systematic mining and ingestion loop for building a coherent AI agent skill library at scale.
[Architecture]
[Skills]
Progressive Disclosure: CLI Windows into Agent Systems_
Agent systems are opaque by default. Progressive disclosure gives operators layered CLI views from quick status checks to full agent internals and decision traces.
[Architecture]
[CLI]
[Tooling]
Ground Truths: Anchoring AI Agents to Reality_
AI agents hallucinate confidently. Ground truths are versioned, scoped facts that anchor agent behavior to reality. Here is how we built and enforce them.
[Architecture]
[Grounding]
Orkestra: Orchestrating AI Agents at Scale_
One orchestrator routes work to specialist AI agents, each with distinct skills and memory. How Orkestra coordinates 46 agents and 466 skills in production.
[Architecture]
[Open Source]
Stay in the loop
That's the full archive_
Every dispatch lands here first. New ones ship straight to the changelog, or subscribe via RSS to get notified the moment one drops.
[Changelog] [RSS]