
1 / 2
【Overview】 Blume is a tool for developers who run coding agents and need a clear record of what those agents actually did. It follows every agent session from the first instruction through to the merged commit, so the work is not lost in a stream of terminal output. Instead of piecing together what happened, you get a structured view of each change and the evidence behind it. The core problem Blume solves is review. Coding agents can plan, build and test, but a human still has to decide whether the result is correct before it ships. Blume groups a product change together with its brief, task tree, agent runs, code revision, tests and screenshots as one delivery. Every agent claim is pinned to a specific task, delivery and commit, so you can check test steps and screenshots and approve the exact revision that goes out. Blume is built for developers running agents such as Codex or Claude Code, and for teams that need to review agent-produced code before shipping. It connects to agent harnesses through a plugin, keeps conversation transcripts locally in a SQLite database, and can show usage and spending pulled from providers such as Anthropic, ChatGPT and Cursor. 【Key features】 • Agent session tracking: Blume follows every coding agent session and records each change from the first instruction to the merged commit. • Delivery records: A product change is grouped with its brief, task tree, agent runs, code revision, tests and screenshots as a single delivery. • Evidence-based review: Each agent claim is pinned to a specific task, delivery and commit, so test steps and screenshots can be reviewed before approval. • Agent plugins/harness support: Works with Codex, Claude Code, Grok Bot, OpenClaw, Hermes, pi and other harnesses by connecting through a plugin. • Usage and spend tracking: Fetches usage and spending information from providers such as Anthropic, ChatGPT and Cursor and displays it locally. • Cloud improvements suggestions: Optionally sends conversation excerpts through a transient relay to Vercel AI Gateway and Google Gemini to generate titles, summaries and metadata insights. • Local transcript storage: Reads conversation transcripts from supported AI tools on your device and stores them locally in a SQLite database. 【Use cases】 1. Track coding agent sessions end to end, from the first instruction to the merged commit. 2. Review agent test evidence and screenshots before approving a revision. 3. See what work remains before shipping, and monitor AI tool usage and spend locally. 【Pros and cons】 Pros: - Developers who run coding agents get a durable, reviewable record of each session instead of relying on memory or scattered logs. - Teams get a single approval point tied to an exact commit, which makes sign-off on agent-produced code more defensible. - Usage and spend visibility is available locally, so cost awareness does not require sending data to a dashboard. Cons: - Blume does not keep a cloud archive of conversations after processing. 【FAQ】 Q: How does Blume connect to coding agents? A: Blume works with harnesses including Codex, Claude Code, Grok Bot, OpenClaw, Hermes and pi by connecting through a plugin. Q: Where does Blume store conversation transcripts? A: It reads transcripts from supported AI tools on your device and stores them locally in a SQLite database, with no cloud archive kept after processing. Q: Is cloud processing required for Blume's suggestions? A: No. Cloud improvements suggestions are opt-in only, are never used as an automatic fallback, and require zero-data-retention and no-training routing.
WPS AI is an AI assistant built into WPS Office that helps users write documents, generate presentation outlines and…
Baidu Comate is an AI coding assistant built by Baidu on its ERNIE model. It goes beyond simple autocomplete: it…
UniConverter is an AI media conversion and enhancement tool for video, audio, and images. It combines everyday…
6 similar tools to consider instead of Blume