Waveloom
Terminal coding agent built for DeepSeek's cache economics
Start new thread

Waveloom - Terminal coding agent built for DeepSeek's cache economics

by•
为 DeepSeek 前缀缓存定制的终端 Code Agent（纯 Go），缓存命中率 95-99%，输入成本降至 1/50。A terminal coding agent optimized for DeepSeek prefix caching — 95-99% cache hit, 1/50th the cost. - Menfre01/waveloom

Add a comment

Replies

Best
Maker
📌
Hey PH! I'm Menfre, maker of Waveloom. I've been using Claude Code since day one — love the UX, hate the token burn. So I built Waveloom: a terminal-native AI coding agent in Go, with the same TUI quality, but engineered for cost efficiency at scale. The key difference is persistent context caching — a 4-level compaction system (Snip/Prune/Summarize) that keeps cache hits high across long sessions. No more paying to re-read the same context every turn. Also built-in: multi-agent orchestration (Fork, Explore, Cold, verification), native MCP client, plan mode, and a skill system. All in one binary. Open source, written in Go + Bubble Tea. Try it, break it, tell me what sucks.

The 95% cache hit rate is honestly wild for a terminal agent, love seeing pure Go in this space too. One thing that would make me reach for this daily though - adding a small inline cost tracker that shows estimated tokens saved per session, maybe even a running total in the status bar. Would make the 1/50 cost claim feel tangible instead of something you have to take on faith.

A 1/50 cost reduction with DeepSeek caching sounds huge. One thing that would help adoption though is adding a local token usage dashboard so we can actually see the cache hit ratio in real time and confirm the savings match what's claimed.

Nice focus on cutting DeepSeek costs with prefix caching. One thing that would help me adopt it faster is a built-in token and cost dashboard per session, so I can see in real time how much I'm saving versus a regular call, and track it across projects.