Tako, a Knowledge Search API
I'm Bobby, CTO of Tako. We just launched our Knowledge Search API. Our API takes natural-language prompts like "Nvidia M&A history" and returns visual answers and grounding text sourced from real-time, structured data (example: https://trytako.com/card/YHloo1Ea7GRnBr_s5r6s/). Most AI systems can’t effectively reason about real-time, structured data. One reason is access: a lot of the most valuable info is trapped in databases web crawlers can't index. Google solves this with a team of 2k+ engineers that ingest data (stocks, sports, etc) into a proprietary Knowledge…
In plain words
Tako is a Knowledge Search API that accepts natural-language queries and returns visual answers with supporting text from real-time, structured data sources. Designed for developers building AI applications, it enables systems to reason effectively about current information trapped in databases that web crawlers cannot access. The API aims to provide the knowledge search and visualization capabilities typically found in large-scale knowledge graphs, accessible via a developer-friendly interface for augmenting language model functionality.
written from the facts on this page · September 2026
From the sources
In the maker’s words, at launch
I'm Bobby, CTO of Tako. We just launched our Knowledge Search API. Our API takes natural-language prompts like "Nvidia M&A history" and returns visual answers and grounding text sourced from real-time, structured data (example: https://trytako.com/card/YHloo1Ea7GRnBr_s5r6s/). Most AI systems can’t effectively reason about real-time, structured data. One reason is access: a lot of the most valuable info is trapped in databases web crawlers can't index. Google solves this with a team of 2k+ engineers that ingest data (stocks, sports, etc) into a proprietary Knowledge Graph. Our goal is to offer developers the same knowledge search + visualization primitives Google has built, tailored for AI use cases, and delivered via API. We seek to augment LLM’s capabilities, and this means that most of our biggest technical challenges stem from not getting a lot for “free” from LLMs. For example, RAG architectures that generate final outputs with LLMs introduce accuracy issues we can’t tolerate, and are too slow. We’ve built a Generative Augmented Search (GAS) architecture that uses LLMs (currently Llama 3.3-70B on Cerebras) to analyze input queries (~200 ms) but use deterministic retrieval for most output generation. The data in our knowledge graph generally isn’t available in LLMs or the web, so we have to acquire it directly from sources (including licensing it from authoritative providers like S&P Global). A limitation of this approach is that some developers want us to offer the flexibility of LLM analysis across web sources, even if it means tolerating non-authoritative sourcing and some hallucination. We’re working on some solutions to that now. Curious to hear how other people are fighting hallucination. I'd love your feedback on the product (and happy to discuss/answer questions about it/the tech stack)
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 16d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 26d ago · cactuscompute.com
Launched alongside, May 2025
the whole month →
- C9
Life & fun · 2025 · felixrieseberg.github.io
