EVAL Engine

Blog

Latest news and updates from our team

Bar chart: Decision-4B (76.39%) and Decision-0.8B (62.61%) mean accuracy against other decision models across 2,800 historical cases

Decision-4B and Decision-0.8B: small models that choose instead of chat

Open-weight LoRA models on Qwen3.5 that pick one listed option in a single forward pass, with probabilities, small enough to run on a device.

Unbound homepage: 'A model with no guardrails. In your browser.' with Chat, Generate Images and Download Model buttons

Unbound: small uncensored models that run on your phone and in your browser

How we built the Unbound E2B, E4B and Q-0.8B models, how we scored them, and the iOS app and browser client that run them locally.

Decenchro homepage hero: 'Deploy secure and auditable AI Agents' beside an 'Actions on the record' chart

Decenchro: an AI agent with an on-chain memory and audit trail

Decenchro runs the Hermes agent with memory stored on Chromia, a guard on risky tool calls, and an ERC-8004 identity on BNB Smart Chain.

EVAL Engine: Running a Multica agent runtime on Render

Running a Multica agent runtime on Render

Packaging the Multica daemon and Claude Code into one Docker image that runs as a Render background worker.

Plush.fun homepage: 'Unbox instantly on-chain. Own it in real life!' above the AI Plushie Genesis Collection card

PlushFun: blind-box checkout, fair draws and talking plushies

How the PlushFun storefront handles multi-method checkout, idempotent CHR payments, verifiable draws, and the audio pipeline for the plush itself.

EVAL Engine: Orchestrator: turning a plain request into scoped, dispatched agent work

Orchestrator: turning a plain request into scoped, dispatched agent work

A repo of reusable skills that lets coding agents scope work, write tickets, and dispatch and verify tasks through Multica.

EVAL Engine: A design system built for AI agents to read

A design system built for AI agents to read

Brand kits, agent skills and video projects in one repo, so a coding agent can make on-brand pages, decks and videos from a one-line prompt.

EVAL Engine: Connecting OpenClaw agents to XiaoZhi voice devices

Connecting OpenClaw agents to XiaoZhi voice devices

A two-way bridge that lets OpenClaw agents speak through, control and query XiaoZhi ESP32 devices, plus HTTP endpoints for pushing speech.

EVAL Engine: Marketing Claw: a containerized Claude agent for Chromia's Telegram

Marketing Claw: a containerized Claude agent for Chromia's Telegram

A fork of NanoClaw that runs Claude agents in containers, adds Telegram, Chromia's MCP server, web search and video tools for marketing.

EVAL Engine: An agent that archives tweets to Chromia FileHub, paid per job

An agent that archives tweets to Chromia FileHub, paid per job

A small seller agent on Virtuals ACP and an x402 endpoint that screenshot a tweet and store the image and metadata on Chromia FileHub.

Chromia Vault deposit success screen showing 7777 EVAL bridged from Base Sepolia to Chromia Token Chain

Bridging an ERC-20 between Base and Chromia, step by step

A walkthrough for deploying a test ERC-20 on Base and registering a token-bridge mapping on a Chromia token chain with Hardhat and chr.

EVAL Engine: Driving a UBTECH AlphaMini robot with an LLM agent

Driving a UBTECH AlphaMini robot with an LLM agent

A small Python demo that gives an AlphaMini humanoid a tool-calling LLM agent, conversation memory and an HTTP endpoint.

EVAL Engine: ChromBot: what we built on top of open-source XiaoZhi

ChromBot: what we built on top of open-source XiaoZhi

Our firmware, server and MCP additions to the open-source XiaoZhi ESP32 voice assistant, which together became the ChromBot companion.

EVAL Engine: Selling over HTTP 402: our x402 pre-order service

Selling over HTTP 402: our x402 pre-order service

A Bun and Express service that takes paid ChromBot pre-order registrations on Base using x402, plus a self-hosted facilitator.

Chrombot's face: two glowing white eyes on a dark CRT-style screen

ChromBot: an ESP32 AI companion with a Chromia account

ChromBot is an ESP32 voice companion. Its web app handles Bluetooth setup and login, and a Chromia contract stores device activity summaries.

EVAL Engine: A Chromia vector store for LangChain.js

A Chromia vector store for LangChain.js

Our Chromia vector store integration for LangChain.js, built on postchain-client, in our fork of langchain-ai/langchainjs.

Vector DB agent sandbox UI with a Crypto Research Agent input panel and a description of its embedding pipeline

A browser sandbox for retrieval-backed agents

A Next.js sandbox for ingesting text into a vector database and querying it with retrieval, plus a set of agent pattern implementations.

EVAL Engine: An MCP server for Kaito Yaps attention scores

An MCP server for Kaito Yaps attention scores

A small unofficial MCP server that lets Claude Desktop or Cursor look up Kaito Yaps attention scores for X accounts.

EVAL Engine: DeFAI eval: testing crypto knowledge and exact address extraction

DeFAI eval: testing crypto knowledge and exact address extraction

A 119-question benchmark that checks whether LLMs know crypto basics and can pull wallet addresses and tx hashes out of noisy text.

EVAL Engine Tweet Evaluation Report showing overall, truth, accuracy, creativity and engagement scores

EVAL Engine v1: scoring agent replies and recording them on Chromia

How the first EVAL Engine scored AI agents' X replies with four LLM judges and recorded every result on Chromia through a Rell dapp.