Local-first sidecar for monitoring coding agents, their rules and skills, usage, drift and proposed configuration improvements.
Blume is a desktop sidecar for people who work with coding agents. The current app centralizes live agent status, hidden rules/skills/hooks, provider usage and configuration files across supported harnesses including Cursor, Claude Code, Codex, omp and Pi. Its Improve workflow analyzes recurring corrections and patterns, shows the evidence and exact diff, and keeps the user in control of applying or dismissing a suggested change. Some broader automation and analytics capabilities are still explicitly marked Soon or on the horizon, so they are not counted as current capability. 12
Blume is currently free of charge. Its terms explicitly say paid features may be introduced later, so the present free model should not be treated as a permanent pricing guarantee. Current first-party material says the app runs on macOS, Windows and Linux. That gives Blume unusually low evaluation friction for a developer tool that spans multiple agent ecosystems, although there is no current paid-plan structure to judge for long-term price predictability. 32
Privacy architecture is one of Blume's clearest differentiators. Conversation transcripts are read from supported tools and stored locally in SQLite. Optional cloud-powered improvement analysis is user-selected rather than an automatic fallback; the privacy page says relevant excerpts may then pass through Blume's transient processing path and configured model providers, while conversation content is not retained as a long-term Blume cloud archive. The fixed 50-source panel produced no eligible exact-product numeric evidence. Outside that panel, recent launch coverage and one independent review corroborate the product surface, but a single 3.7/5 review is not enough to establish a robust external rating class and is therefore retained qualitatively rather than blended into the score. 1456
Blume is most useful for developers who repeatedly move between coding agents and accumulate scattered rules, skills, hooks and provider limits. It addresses the operational layer around agentic coding rather than replacing the coding agents themselves: the value comes from seeing what each agent is doing, understanding the context shaping it and turning recurring corrections into maintainable project guidance. Developers who use only one agent occasionally, or who do not want a desktop sidecar reading local harness data, will get less value. 14
Blume's strongest combination is breadth plus local-first control: multiple coding harnesses, session visibility, quota tracking, configuration inspection and approval-based improvements are gathered in one desktop surface while primary conversation history remains local. Current free access and cross-platform availability add substantial value. The
age and evidence maturity. Several ambitious features remain unfinished, exact-product independent validation is still sparse, and optional cloud processing requires a different privacy trade-off from the default local path. The 8.5 candidate score therefore carries LOW confidence and is not rank-eligible. 14356
Products matched by shared tasks, audience, workflow, product type and capabilities.