Codex context bloat? 87% avg reduction on SWE-bench Verified traces
Get picks like this daily. Today's launch numbers, the top launches, AI and tech news, and a daily opportunity pick, straight to your inbox.
Want to reach these readers? Sponsor an issue.
If you had to build a context window manager in 24h, would you stick to the existing model or come up with something better?Here's what I did:1. Built a proxy that intercepts Codex's calls to OpenAI and rewrites them on the fly.2. Replayed 3,807 rounds of SWE-bench Verified traces through it: avg prompt 44k → 6k tokens (-87%).3. Posted it to HN to get the next reduction applied to my confidence interval — starting with the inevitable "How about accuracy?"npx -y pando-proxy · github.com/human-software-us/pando-proxy
HN · launched Apr 24, 2026 · 10 upvotes (top 41%) · 0 views
Niche: llm optimization tools · 127 launches, 9 in the last 7 days ▲ +50% · Explore this niche →
Category and tags
- Horizontal
- Dev tools
- Developer
- AI-native
- Commercial product
Visit website ↗ Source listing ↗
More like this: all 127 launches in llm optimization tools →
Could you build this?
Partial The proxy server itself is simple, but achieving an 87% context reduction without degrading code benchmark performance requires sophisticated prompt deduplication, chunking heuristics, and active memory management.
What it would actually take: An HTTP/reverse-proxy built in Node.js or Go that intercepts OpenAI Responses/Completions payloads, tokenizes request bodies, tracks conversation DAGs, and applies stateful diffing and retrieval functions. The hardest part is the semantic chunking and active memory lifecycle algorithm that decides what context to drop or recall without hallucination or broken dependencies. Requires deep expertise in LLM context manipulation, SWE-bench evaluation methodology, and stateful proxy infrastructure.
Discussion
2 comments analyzed.
Concerns raised: Proxy overhead makes short sessions more expensive than direct API calls, Two extra model calls per round (chunker + working_memory_update) add hidden costs, Unclear total cost comparison when factoring in multiple model invocations
Competitors
Other products that read as similar to this one : 114 launches clear the similarity bar, closest 8 shown.
Attention rank: #58 of 115 (itself plus its competitors, highest first, normalized so YC and Product Hunt are compared fairly).
Launched 171 days after the earliest competitor.
- Edgee Codex Compressor V2 · ph · 2026-09-18 · 90 upvotes · similarity 0.46
- Edgee Codex Compressor · ph · 2026-04-12 · 166 upvotes · similarity 0.45
- Swift-Qwen3.8-27B, -58.3% thinking, x1.95 speed, accuracy of xhigh · hn · 2026-09-16 · 30 upvotes · similarity 0.41
- codex_state_monitor · github · 2026-09-20 · 13 upvotes · similarity 0.40
- Context Gateway · hn · 2026-03-13 · 97 upvotes · similarity 0.40
- RealtimeVAD · hn · 2026-11-12 · 33 upvotes · similarity 0.40
- Token compression CLI to save Codex/Astra costs · hn · 2026-09-30 · 11 upvotes · similarity 0.40
- yoshi · github · 2026-09-18 · 22 upvotes · similarity 0.39
The niche, labels, vibe code rating and competitor comparison on this page are analysis by Nicheloom. If you use them, please credit nicheloom.com. How you may use our data.