How Blume codebase improvements work - Blume Blog

Arrow Left<br>Back to blog<br>How Blume codebase improvements work<br>Blume Team Calendar June 23, 2026 5 min read

How Improvements Work

Blume watches your coding-agent conversations and quietly looks for friction —<br>the questions your agent kept misunderstanding, the rule you keep re-explaining,<br>the config that never quite matched your setup. When the same problem shows up<br>often enough, Blume drafts a concrete fix and shows it to you. You review the<br>diff and decide whether to apply it.

This article explains what the feature does, how to turn it on, and what happens<br>under the hood.

The idea

Friction is hard to act on in the moment. A single confusing exchange isn't<br>worth changing your setup over — but the third time the agent makes the same<br>wrong assumption, there's a real gap worth closing. The trouble is that you<br>rarely notice the pattern, because each instance is buried in a different chat.

Improvements close that loop:

Watch your conversations for repeated signals of friction.

Cluster similar signals across chats, so a one-off doesn't trigger noise.

Propose a specific change — a new rule, hook, skill, or doc — once a<br>cluster is backed by enough evidence.

Review and apply on your terms, with a full diff and one-click undo.

Two things matter most: it only acts on repeated evidence, and it runs<br>entirely on your machine . The analysis goes through your own agent CLI<br>(Claude Code, Codex, or Cursor) using your existing plan's quota. Nothing about<br>your conversations is uploaded to Blume or any third party.

Turning it on

Improvements are off by default — it's opt-in. In the Improve row at the top<br>of the suggestions list, click Configure . You'll choose two things:

When to run

Smart timing (recommended) — runs at the end of a session, in the last 15<br>minutes of your plan's quota window, but only while more than 25% of that<br>window's quota is still unused. In other words, it spends quota you were about<br>to lose anyway, and stays out of your way the rest of the time.

When chats go quiet — analyzes each conversation as soon as it goes idle.<br>This gives you the fastest feedback, at the cost of using quota throughout the<br>day.

Which plan to use

Pick the connected plan (Claude Code, Codex, Cursor) that should run the<br>analysis. The list is populated from your connected plans. If you don't choose,<br>Blume picks one for you.

Click Turn on , and the scheduler starts. From then on, the Improve row shows<br>On · runs locally .

Tuning what it does

Once enabled, Configure opens the full settings page. Everything here is<br>opt-out — sensible defaults are already selected, and you narrow things down.

Harnesses to improve. A checklist of your installed agent CLIs. Checked<br>harnesses get improved; uncheck one to skip it. New harnesses you install later<br>are improved by default.

Scopes — where changes are allowed to land:

Project — files in the current repository (e.g. .github/AGENTS.md).

Global — your machine-wide config (e.g. ~/.claude/settings.json).

Both are on by default.

Reviewing a suggestion

Each suggestion appears as a card in the list:

A title describing the change (e.g. "Add Claude Code preference for JSON<br>responses").

An evidence badge — "came up 3 times in your chats" — so you can see it's<br>grounded in a real pattern, not a guess.

One or more operation rows , each showing the action (+ create, ~<br>update, − remove), the artifact type, a short summary, and where it lands<br>(project or global, which harness it applies to).

Click Preview to see the full picture: a unified diff per file, the absolute<br>path it touches, and the expected effect. From the preview you can:

See evidence — the raw conversation excerpts that motivated the change.

Snooze or Dismiss if it's not useful right now.

Apply — write the changes to disk.

Applied, dismissed, and snoozed suggestions move to resolved , where applied<br>ones can be undone .

Safe by construction

Applying is all-or-nothing . Before writing anything, Blume verifies every<br>target: a create must not already exist, and an update or remove must match<br>the file exactly as it was when the suggestion was drafted. If anything has<br>changed underneath it, nothing is written and the suggestion is flagged for<br>review instead of silently clobbering your work.

Each applied change records what it replaced, so Undo can restore the<br>previous contents — again, only if the file still matches what Blume wrote.

What happens under the hood

When a conversation is analyzed, it flows through a local pipeline. Each stage<br>that needs judgment calls your agent CLI; the rest is plain local logic.

Conversation<br>├─ Extraction agent reads the transcript, pulls out "signals"<br>│ (friction points, with evidence and a pain score)<br>├─ Clustering agent groups new signals with similar ones from<br>│ past conversations — or starts a new cluster<br>├─ Promotion local thresholds decide when a cluster has enough<br>│ evidence (signals × conversations × pain) to act on<br>├─ Planning...