Months of Claude Code sessions you thought were gone.

The problem

vir is not a memory layer your agent writes to. It reads what already happened.

How it works

A loop, not a pipeline.

Sessions become notes. Notes feed the next session. The daemon keeps it turning while you work.

What it writes

Anatomy of a note.

Every memory an agent-memory store keeps is for the agent. This is for you — and the agent reads it too.

You won't read four hundred of these, and you're not meant to. The agent reads them, through MCP. You read the three that matter, on the day they matter, because vir query found them or Obsidian's graph put one next to what you were already writing.

---
topic: "kie-200-error-body"
aliases:
  - "kie-200-error-body"
category: gotcha
project: "growthq"
session_id: 4f2a9c31-7d0e-4b8a-9c55-1e2f3a4b5c6d
date: 2026-06-01T09:14:22.000Z
confidence: 0.86
themes:
  - "kie-error-handling"
  - "retry-safety"
branches:
  - "fix/blank-posts"
---
Project: [[growthq]]
Category: [[gotcha]]

## Summary

This session debugged blank posts in growthq's image pipeline. The Kie.ai endpoint answers HTTP 200 even when generation fails: the error is only visible as `{ "code": 422, "msg": "..." }` in the body, so `res.ok` treats every failure as success and the job polls forever.

## What Was Learned

- **Check the body's `code`, not the HTTP status.** The client now throws when `code !== 200` before trusting `res.ok`.
- **Retry only on `code` 5xx.** 422 means the prompt was rejected and will be rejected again.

## Context

The generation queue had been producing blank posts for two days.

## Related

- [[retry-with-backoff-0c91be7a|retry-with-backoff]]
- [[kie-ai-task-polling-loop-a83e21f0|kie-ai-task-polling-loop]]

Inputs

Three inputs, one vault.

• Agent sessions

  Claude Code and Codex. Retroactive: months of existing transcripts become notes in one run — and new sessions keep flowing in.

• Web articles

  Clip a page to a folder, for example with Obsidian Web Clipper. vir distills it into the same vault.

• PDFs and papers

  Drop a paper in. It gets read, summarized, and linked next to the sessions that cited it.

Everything embeds into one vector space (built-in or Ollama, TF-IDF fallback), so vir query searches across all three.

Retrieval

Ask the vault. Or let the agent ask it.

$ vir query "how did we handle kie 200 errors" Kie.ai answers HTTP 200 even whengeneration fails; the failure only showsas code 422 in the body. Parse the bodyfirst, throw on code !== 200, and retryonly on 5xx. ────────────────────────────────────────  · gotchas/kie-200-error-body-4f2a9c31  0.81  · patterns/retry-with-backoff-0c91be7a  0.64────────────────────────────────────────sources 2 · via embedding · searched 460

Your numbers

Run this before you install anything.

Mine are one machine's. Yours are the only ones that matter, and you already have them — Claude Code has been writing transcripts to your disk since the day you started.

1. First number

   Every session Claude Code still has on disk. vir reads all of them in one pass.

2. Second number

   The ones older than 25 days. Claude Code prunes at around 30, so that's roughly what disappears in the next five — whether or not you do anything.

Both commands only count files. They read nothing, send nothing, and change nothing. Not all of the first number becomes notes — most transcripts are subagent runs and workflow phases, and vir skips those.

What it costs you

Time and money, up front.

1. 60 secondsnpm install -g, then vir init — a wizard: provider, models, vault path.Free.
2. One runMonths of history become a vault.$25.85 for 297 sessions over six months, on the Anthropic API. On a Claude subscription via the claude-cli provider: no API spend at all.
3. Ongoingvir schedule install registers a daemon that keeps the vault current.Median $0.004 per session, $0.14 at the 90th percentile.

vir run --dry-run estimates your own history before you commit to a full run; vir cost reports what you actually spent.

Claude Code forgets. Your vault doesn't.

Install.

vir init is a wizard: provider, models, vault path. vir run does one pass over your sessions and writes notes. When you like the output, vir schedule install registers a daemon that keeps the vault current.

Full setup, configuration, and every command: the docs.

macOS or Linux · Node 20+ · Claude Code or Codex · Obsidian optional — the output is plain markdown either way