Impact-Site-Verification: 1e78237b-3fa7-4806-bddb-9f005f43c040
directree-verify=901579de7df1146af265cf42401d3f68
Wonderlaunchwonderlaunch

Blume.codes

A desktop sidecar that monitors coding agents and improves their context

The Problem

Developers juggling multiple coding agents lose track of hidden rules, configuration drift, and token usage, leading to unexpected behavior and wasted resources.

Overview

Blume Sidecar is a desktop sidecar that visualizes and manages coding agents running on a developer’s machine. The app attaches to each running agent, reads the internal graph that defines its rules, skills, hooks, and any hidden configuration files. It presents a narrow window that lists every agent, its current state (idle, working, finished) and the exact instructions guiding it. All conversation logs are written to a local database on the developer’s device, keeping data private and avoiding cloud sync. When an agent proposes a change, Blume captures the diff, shows the exact code modifications and the supporting evidence, then lets the user approve, dismiss, or stash the suggestion. It also monitors token consumption and plan limits for Claude Code and Codex, warning the user before limits are exceeded. Future updates plan auto‑fix detection for mismatched setups, analytics on agent performance, and shared domain models for team collaboration, but the current release focuses on visibility, local storage, and manual approval workflow.

Feedback

Have feedback for the maker?

Sign in to leave a review, report a bug, or suggest a feature.

Visit Blume.codes

Submitted by

Peder Aaby

Peder Aaby

@blumedotcodes
Launch DateComing Soon
CategoryAI Agents & Automation
PlatformmacOS, Windows, Linux
Pricingfree

Spread the word

Help this maker reach the front page!

← Back

Blume.codes

A desktop sidecar that monitors coding agents and improves their context

🔥Launching In

00
HRS
00
MIN
00
SEC
Peder Aaby

Meet the Maker

Peder Aaby

Follow on X

Overview

Blume Sidecar is a desktop sidecar that visualizes and manages coding agents running on a developer’s machine. The app attaches to each running agent, reads the internal graph that defines its rules, skills, hooks, and any hidden configuration files. It presents a narrow window that lists every agent, its current state (idle, working, finished) and the exact instructions guiding it. All conversation logs are written to a local database on the developer’s device, keeping data private and avoiding cloud sync. When an agent proposes a change, Blume captures the diff, shows the exact code modifications and the supporting evidence, then lets the user approve, dismiss, or stash the suggestion. It also monitors token consumption and plan limits for Claude Code and Codex, warning the user before limits are exceeded. Future updates plan auto‑fix detection for mismatched setups, analytics on agent performance, and shared domain models for team collaboration, but the current release focuses on visibility, local storage, and manual approval workflow.

The Vision

Developers juggling multiple coding agents lose track of hidden rules, configuration drift, and token usage, leading to unexpected behavior and wasted resources.