zkAGI · Case studies
Every project below runs on the Zynapse API, one OpenAI-compatible endpoint with the privacy level chosen per call, and was delivered by a forward-deployed engineer who stayed on the account. Custom APIs, tight deadlines, models configured at the privacy level each customer needed.
SDK Infinity
PawPadNEOBOLICALumen Estates BY ZKAGI
How zkAGI rebuilt Wandrlust's multimodal agent stack, cut P95 latency, simulated 1M users offline with Mirofish, and shipped GPS privacy by default. Concept by Catchfire AI.
Read the case study →zkAGI built SDK Infinity a phone-first company brain and an agent pipeline that turns a voice note into a live ERP demo in about 30 minutes. SDK Infinity is zkAGI's ERP channel partner in India.
Read the case study →Arbis.ai runs its agent governance reviewer on Zynapse's mixture-of-agents through a sealed TEE. zkAGI built real token streaming, off-path attestation checks and a judge mode to meet Arbis's latency and privacy needs. Arbis is zkAGI's channel partner in New York.
Read the case study →zkAGI built BRAT, a vision-and-voice agent that rates your outfit and roasts it, for RAMM.ai in New York. Camera frames are analysed on Zynapse, the verdict is spoken, and nothing of the user's face is kept.
Read the case study →zkAGI built IdolMaker AI for Beatfarm in Boston: upload a song, get a K-pop style music video with a virtual idol in 5 to 15 minutes. Beat detection, mood analysis and scene planning on Zynapse, rendering on zkAGI's video engine.
Read the case study →
PawPadPawPad, a zkAGI ecosystem project funded by an Oasis grant, runs trading agents whose signing keys are split 2-of-2 between the user's device and a TEE share on Oasis Sapphire. Staging phase complete.
Read the case study →Neobolica is a neoclassical metal band whose music is written and played by humans, while the music videos and the marketing engine run on zkAGI's video engine and agents.
Read the case study →Lumen Estates, by zkAGI, turns listing photos into cinematic property films in 72 hours, in Listing, Signature and Estate formats, for Singapore and Switzerland. Built on zkAGI's video engine and Zynapse.
Read the case study →From 2019 to 2022 Suraj Venkat set up the forward-deployed engineering and solutions practice at AI2Easy, formerly Tektorch AI, installing AI for go-to-market teams and closing multi-year six-figure deals before founding zkAGI.
Read the case study →One OpenAI-compatible endpoint. Privacy per call: zero-data-retention providers, a sealed TEE lane, or your own GPUs. Mixture-of-agents, PII scrubbing, caps per key, a receipt for every call.
A video agent we use ourselves: MiniMax H3 video generation optimised for our fleet, scenes cut to a timeline, text-to-speech narration, LTX upscaling, a human-driven creative process and evals at every stage. Behind Beatfarm, Neobolica and Lumen Estates.
The same code, a different model name. Privacy is a per-call configuration, and every call returns a receipt.
| Lane | What it does | This project |
|---|---|---|
| zynapse-fast | Zero-data-retention providers via Engy. Fastest, lowest cost. PII scrubbed before the model, rehydrated after. | available |
| zynapse-balanced | Mixture-of-agents cascade on the fast lane: one strong call, a referee, escalation only on disagreement. | available |
| zynapse-tee | Sealed confidential VM on Phala. Hardware attestation checked on every call; receipt carries the attestation verdict. | available |
| self-hosted / on-prem | Open-weight models on the customer's own GPUs (Odoo-style installs, partner nodes). Nothing leaves the building. | available |
Open the customer and partner map →
Custom API, configured to your privacy level, with an engineer who stays until the number moves.
One call to find out whether it fits.
Talk to the FDESee the lanes