Alternatives
Products that do what Codex context bloat? 87% avg reduction on SWE-bench Verified traces does
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
- 1
- 2
- 3
- 4BA
Hi HN, Erik here. Today we launch Butter, an OpenAI-compatible API proxy that caches LLM generations and serves them deterministically on revisit. Since April, we’ve been working on this concept of “muscle memory,” or deterministic replay, for agent systems performing automations. You may recall our first post in May, launching a python package called Muscle Mem: https://news.ycombinator.com/item?id=43988381 Since then, the product has evolved entirely, now taking the form of an LLM Proxy. For a deep dive into this process, check out:…
Oct 2025 · docs.butter.dev
- 5LP
Hi HN, not sure if anyone would be interested, but just wanted to share that I've been maintaining my small tool called 'lowfat' that helps me filters some of my verbose CLI output. It's a single binary, works as an agent hook or a shell wrapper. It has a plugin system to customize filters per command. The idea is pretty simple: agents don't need the full kubectl get -o yaml or any 10k-line dump to make decisions. So that lowfat sits in between, strips the noise, and passes through what matters. Here's my real report after 2 months of personal use: lowfat history --all lowfat plugin…
Jun 2026 · github.com
- 6
- 7
- 8CM
Every MCP tool call dumps raw data into Claude Code's 200K context window. A Playwright snapshot costs 56 KB, 20 GitHub issues cost 59 KB. After 30 minutes, 40% of your context is gone. I built an MCP server that sits between Claude Code and these outputs. It processes them in sandboxes and only returns summaries. 315 KB becomes 5.4 KB. It supports 10 language runtimes, SQLite FTS5 with BM25 ranking for search, and batch execution. Session time before slowdown goes from ~30 min to ~3 hours. MIT licensed, single command install: /plugin marketplace add mksglu/claude-context-mode…
Feb 2026 · github.com
- 9
Ask Claude Code where your usage went. Token audit, limit diagnosis and usage forensics — built from the session logs already on your machine, nothing leaves it. - kelviq/tare
10d ago · github.com
- 10OA
Hi HN, we're Kiran and Vijay! Over the past two years, we have built a columnar storage engine for observability: logs, metrics, and traces. Today, it's exciting for us to show what we've built on top of that foundation: LLM Agent Observability. Given how non-deterministic agents are, storing all traces without sampling was critical for us. But these traces tend to be in the MBs, sometimes GBs - we needed to store them inexpensively. We also needed the queries and analyses to be fast. To meet both these goals, we store them in S3 in our own parquet-like file format, and query them using AWS…
Jul 2026 · oodle.ai
- 11LE
I started using Claude Code (claude --dangerously-skip-permissions) and Codex (codex --yolo) and realized I had no reliable way to know what they actually did. The agent's own output tells you a story, but it's the agent's story. logira records exec, file, and network events at the OS level via eBPF, scoped per run. Events are saved locally in JSONL and SQLite. It ships with default detection rules for credential access, persistence changes, suspicious exec patterns, and more. Observe-only – it never blocks. https://github.com/melonattacker/logira
Mar 2026 · github.com
- 12BA
Hi HN! I'm Erik. We built Butter, an LLM proxy that makes agent systems deterministic by caching and replaying responses, so automations behave consistently across runs. - It’s a chat completions compatible endpoint, making it easy to drop into existing agents with a custom base_url - The cache is template-aware, meaning lookups can treat dynamic content (names, addresses, etc.) as variables You can see it in action in this demo where it memorizes tic-tac-toe games: https://www.youtube.com/watch?v=PWbyeZwPjuY Why we built this: before Butter, we were Pig.dev (YC W25), where we…
Oct 2025 · butter.dev
- 13WB
Hey HN, We’re two developers (co-founders) with a team of 20 who got tired of spending hours reviewing PRs, so we built Infinitcode.ai, an AI-powered code reviewer that: - *Summarizes PRs in plain English*: No more deciphering 1,000-line diff jungles - *Catches more than bugs*: Security holes, performance pitfalls, code smells, even typos (yes, we’ll flag “vurnerabilities” and vulnerabilities) - *Zero onboarding*: Works instantly—no “let me learn your codebase for weeks” nonsense. Why we’re posting: We’re in alpha and need brutal honesty. Roast our tool, mock our UI, or tell us why AI will…
2025 · infinitcode.ai
- 14AO
This is a small PoC Python project for web server access logs analyzing to classify and dynamically block bad bots, such as L7 (application-level) DDoS bots, web scrappers and so on. We'll be happy to gather initial feedback on usability and features, especialy from people having good or bad experience wit bots. *Requirements* The analyzer relies on 3 Tempesta FW specific features which you still can get with other HTTP servers or accelerators: 1. JA5 client fingerprinting (https://tempesta-tech.com/knowledge-base/Traffic-Filtering-b...). This is a HTTP and TLS layers…
Oct 2025 · github.com
- 15AA
Hi HN! This is Salman, Adil, Shuguang and Co working on ArchGW[1] - an open-source lightweight proxy server for prompts - written in Rust and built on top of Envoy[2]. Arch moves the critical but pesky handling and processing of prompts: task understanding, prompt routing, safety, and observability - outside business logic. Its an edge and egress proxy for agentic apps. We've talked to 100s of developers at places like Twilio, GE Healthcare, Redhat, Square, etc and there was a consistent theme in building AI apps: to move past a nascent demo they are left to their own devices in building out…
2025 · github.com
- 16TI
I'm an "ideas person" who messes around with AI on a low budget. I got tired of watching my tokens vanish and context windows filling up while agents fumbled around trying to find the right thing. Agents don't flail like they used to with shell tools, but there are still weak/blind spots and back-and-forth episodes — especially when using tools in combination/sequence. So I built "tilth" today. Or rather, AI built it — every line is Opus 4.6. I spent a lot of my precious tokens getting it to "not shit" (at least several of the different vendors' AI overlords assure me it's not…
Feb 2026 · github.com
- 17AW
I've been presenting at local meetups about Context Engineering, RAG, Skills, etc.. I even have a vbrownbag coming up on LinkedIn about this topic so I figured I would make a basic example that uses bedrock so I can use it in my talks or vbrownbags. Hopefully it's useful.
Apr 2026 · github.com
- 18
- 19AL
Raymond here from Butter.dev, an LLM response cache built as a chat-completions proxy. Today we're launching a key feature for the platform: the ability to generalize on dynamic, templated inputs. Caching at the HTTP request level has the obvious problem of generalizability. Nearly no request is identical, due to templated variables (like names) and metadata (like timestamps), so exact-match cache lookups rarely hit. We solve this at Butter by using LLMs to detect dynamic content in requests and derive their inter-relationships, allowing the cache entry to be stored as a template + variables…
Jan 2026 · blog.butter.dev
- 20MY
LLM observability is an absolute must-have for anyone running something in prod (or prod-like). While all the observability startups are great, you're essentially sending all your OpenAI usage history - prompts, generations, chats - to a random third party. So this script deploys a basic proxy in your Azure account, catches all incoming OpenAI requests, stores logs in your own resource group, and comes with visualizations premade (charts, timelines, chat history, cost estimation, etc). Thanks for any thoughts and feedback!
2023 · github.com
- 21PC
Hi HN, I'm Rob. Contextify indexes every Claude Code and Codex session on your machine into one local, searchable database. The current session, in either tool, can search all of it: /total-recall in Claude Code, $total-recall in Codex. Demo: https://www.youtube.com/watch?v=FvrvRGp4C9M | Mac app: https://contextify.sh (App Store or DMG) | Linux: CLI with a one-line installer, same search. No signup. I split my work between Claude Code and Codex. When I burn through rate limits on one, I switch to the other, and new models keep leapfrogging each other, so the…
Jul 2026 · contextify.sh
- 22IN
Tl;dr: I trained a classifier to route to the least expensive model and reasoning depth to complete the request. Coupling that with additional automated token efficiency techniques has yielded 3x usage for the same spend. For anyone interested in trying it themselves: https://nerfguard.com Various teammates and I switched over to Codex from Claude Code recently. We still bounce between the tools, but Codex’s speed and steerability coupled with performance gains were hard to ignore. One of the downsides was that the per token pricing kicked in way sooner. This is happening across…
Jun 2026
- 23AA
At my last company, we connected Claude Code and Cursor to almost all our internal services via MCP. It made the team incredibly fast, but we hit a wall: permissions. If you give an agent "Read Only" access, it can’t actually fix anything. If you give it "Write" access, it’s only a matter of time before a hallucination or a bad prompt results in a deleted database or a nuked production bucket. We had a few "close calls" that convinced us that simply reducing IAM permissions makes agents useless. I built Armour (https://github.com/fuushyn/armour) to solve this. It’s a…
Jan 2026 · github.com
- 24IB
The only way to go fast is full YOLO mode in your coding agent. I've got the local sandbox figured out (pro tip: Incus VMs work great) but I wanted to keep my agents from doing things like inadvertently blowing up my cloud services or chasing a prompt to POST to some random website. I struggle most with this on my side projects where my permission model isn't quite as robust as it is at the office. I started with a firewall on the Incus container but every time the agent needed access to something new, I was poking more holes in it - and it didn't differentiate between HTTP verbs. I've been…
Jul 2026 · trollbridge.dev
Ranked by how close each launch is in meaning, then by votes. Refine with a description →