We went quiet for a while. We're back. KlaatAI is an AI coding assistant. One router, Klaatu, picks the cheapest model tier that can do each job, across 47 models from 15 labs. It works today in web chat, the KlaatCode CLI and VS Code. The Free plan needs no card, and we're in beta.

Earlier this week one of our hosting provider @Railway had a security incident that may have exposed some of our service keys. As a precaution we paused access, replaced every key and re-checked each service. Klaat AI is back up. Thanks for your patience. Questions? Reply here

We are currently shutting down KlaatAI for all customers. We are afraid @Railway has been compromised. We are digging into it and will get back to you ASAP!

Most coding agents can read your code. The hard part is understanding how your code is connected. We’ve significantly improved KlaatCode’s Code Graph, giving the agent a deeper understanding of relationships between files, symbols, functions, classes, imports, dependencies, and execution paths. Instead of searching thousands of lines blindly, KlaatCode can build a structural picture of the codebase and use it to reason about what actually needs to change. That means: ⚡ Better repository understanding 🧠 More precise code navigation 🔗 Dependency-aware changes 🛠️ Better multi-file edits 🚀 More reliable autonomous coding Code agents shouldn't just search your repository. They should understand it. We wrote about how we're building the next generation of KlaatCode's Code Graph: Read the article → klaatai.com/articles/code-gr… #KlaatAI #KlaatCode #AI #CodingAgents #DevTools #OpenSource

AI assistants shouldn’t reset every time you start a new conversation. Today we’re breaking down how KlaatAI’s cross-session memory works, helping KlaatAI remember what matters across conversations, understand your working style, and become more useful over time. The goal isn’t to remember everything. It’s to remember the right things. 🧠 Persistent context 🔄 Cross-session continuity 👨‍💻 Personalized coding workflows 🔒 User-controlled memory. We wrote about how we’re building it and why memory is becoming a core part of AI agents. Read: klaatai.com/articles/cross-s… AI that gets to know how you work, instead of making you start over.#AI #AIAgents #KlaatAI #Memory #DeveloperTools

Been using this lately and honestly, it’s pretty impressive. 👀 The fact that you can get Kimi K3 + Qwen 3.8 Max on KlaatAI’s free tier right now makes it even more worth trying. If you’re into coding with AI, give KlaatCode a shot. 🚀

This is exactly why we built KlaatCode. 🔥 And yes — you can try it for free. For a limited time, KlaatAI’s free tier includes access to Kimi K3 and Qwen 3.8 Max. Build. Ship. Experiment. 🚀 No expensive setup required. Try KlaatCode → klaatai.com

KlaatCode V2.4.2 is out now. Use DeepSeek 4 Flash for free.

📢Meet Qwen3.8-Max — our most capable model to date. Next week, the open weights of Qwen3.8-Max will be released, and Qwen3.8-27B is also going open-weights to meet you all!🎉 Qwen3.8-Max, a new bar for coding and cowork at 2.4T parameters: - Autonomous coding: 10+ days of self-evolving development, from empty folder to production without hand-holding, complete project trace in the GitHub：github.com/qwen-code-dev-bot… - Real work, real results: Production-quality deliverables across hundreds of professions. - Long-horizon mastery: System-level autonomous planning with closed-loop adaptive learning, driving 500+ turns of chip design optimization and 365 days of e-commerce strategy. - Native multimodal intelligence: Vision isn't just input — it's a continuous feedback loop for planning, execution, and self-correction. 💰Pricing: Input: $2.0 / M tokens Output: $6.0 / M tokens Implicit Caching: $0.25 / M tokens Start building with Qwen3.8-Max! 🚀 📖 Blog: qwen.ai/blog?id=qwen3.8 ✅ Qwen Studio: chat.qwen.ai/?models=qwen3.8… ⚡ API: qwencloud.com/models/qwen3.8…

AI is getting way too expensive" is on the HN front page again. We're a small team building a model router precisely because of this. The thing I keep relearning: most coding-agent spend isn't the hard reasoning. It's everything around it: retries, tool calls, failovers, re-reading context that didn't change. so we made two calls early: route every request to the cheapest model that can actually do the job, and never count tool calls or retries against limits. still tiny. still shipping. But I think cost is the wedge Frontier Labs can't defend.

Cursor shipped a router last week. their claim: frontier-quality results at 60% lower cost by auto-picking the model per request. this is the entire bet behind Klaatai. Your rename doesn't need the same brain as your race condition. differences in how we do it: we route across 30+ models from every provider (not one vendor's lineup), tiers from nano all the way to titan, and the same router runs your chat, CLI, and VS Code. Free tier included. Routing isn't a feature anymore. It's becoming the layer.

Your linter fix doesn't need the same model as your race condition. Klaatu-o1 scores each request and routes it across 30+ frontier models in tiers from nano up to the new Titan. Big models when the task earns it, cheap ones when it doesn't. And tool calls, retries, and failovers never count against your message limits. Free tier, no card → klaatai.com

Paying frontier rates to rename a variable is pure waste. We put Klaat Code head-to-head against Claude Code, Cursor and opencode on the exact same 33 real coding tasks. Same prompts. Same verification. No cherry-picking. Results: → Klaat Code: 33/33 · $0.027/task → Claude Code (Sonnet 5): 33/33 · $0.146/task Same accuracy. 5.4× less money. The only difference: Klaatu-o1 routes every message to the cheapest of 30+ models that can actually handle it. Hard problems get frontier power. Everything else stays cheap. Tool calls are free. Code Graph + memory cuts tokens 5–15×. GitHub already at 314 stars: github.com/KlaatAI/klaatcode Reproduce the entire benchmark yourself in under 2 minutes: git clone github.com/KlaatAI/klaatcode bun install && bun run bench Live on Product Hunt right now: producthunt.com/products/kla… Try it free (no card): klaatai.com/klaatu Poll (reply with number): Still paying full frontier rates for trivial edits Already built your own router About to run the bench and tell us we’re full of shit (or not) Drop the numbers. Roast the methodology. Tell us which task broke your expectations. We answer every reply. Who’s first?👇

🚀 300+ GitHub Stars! One year ago, KlaatAI was just an idea. Today, Klaat Code has crossed 300+ GitHub stars. ⭐ This isn’t just another AI wrapper. We built Klaatu, our intelligent multi-model routing engine that automatically selects the right frontier model for each coding task—balancing reasoning quality, latency, and cost. With Klaat Code you get: ⚡ Intelligent multi-model routing 🧠 Repository-aware coding with Code Graph 💬 Unified chat, terminal & API experience 🚀 Faster, cost-efficient AI coding workflows We recently benchmarked Klaat Code against leading AI coding CLIs and were excited to see it deliver strong coding performance while significantly reducing inference costs through intelligent routing. This milestone wouldn’t have been possible without everyone who: ⭐ Starred the project 🐛 Reported bugs 💡 Shared ideas 🔥 Contributed code ❤️ Believed in what we’re building We’re only getting started. Over the coming months we’re focused on: Better agents More intelligent routing Faster inference More supported models Enterprise features An even better developer experience If you haven’t tried Klaat Code yet, we’d love your feedback. ⭐ GitHub: github.com/klaatai/klaatcode Every issue, feature request, and star helps us build something developers genuinely love. Thank you for being part of this journey. ❤️ #opensource #AI #DeveloperTools #LLM #GitHub #Coding #BuildInPublic #KlaatAI

🚀 Open-source AI coding agent for terminal with Claude Code-grade accuracy. See link below 👇

claude code at 18% of the cost, with a reproducible benchmark to back it klaatcode just hit 174 stars and trending in topic:ai on github here is what it does: > per-request cost routing across 5 tiers (nano > heavy), so simple edits don't burn reasoning tokens > code-graph index instead of grep claims 5-15x fewer tokens per task than whole-file reading > standalone binary, no node/bun runtime at runtime > free tool calls inside a request, only user messages count against quota their bench: same 30 fixtures as the claude code test set, 30/30 solved either way, $0.026/task vs $0.146/task you can rerun it with `bun run bench` and check the math yourself setup: npm i -g klaatcode && klaatcode apache 2.0, npm package is live, repo is in CS the claude code tax is optional now

🚀 200+ GitHub Stars! Thank you to everyone supporting Klaat Code ❤️ Open-source AI coding agent. ✅ 33/33 benchmark tasks 💰 $0.89 total cost ⚡ ~5.4× cheaper than Claude Code (same benchmark) Star the repo & try it yourself ⭐ github.com/klaatai/klaatcode

People said, “Run Composer 2.5 Standard it’ll beat you on cost.” We did. ✅ Composer 2.5: 33/33 — ~$1.54 — 113s median/task ✅ Klaat Code: 33/33 — $0.89 — 23s median/task The benchmark is fully open source. Verify it yourself. klaatai.com/benchmarks