ASI:One Mini | ASI:One Documentation

Efficient Reasoning for Everyday Agent Workflows

Introduction

ASI:One Mini is a streamlined reasoning model engineered for everyday agent discovery and routine automation tasks. It focuses on speed, simplicity, and resource-frugal deployment—the perfect engine for chatbots, scheduling agents, customer-support flows, and light on-chain utilities.

Performance Highlights

Metric	ASI:One Mini
MMLU Benchmark	85 %
Context Window	32 k tokens

Why Choose ASI 1-Mini?

Fast, Low-Friction Deployment

Runs comfortably on mid-range GPUs or modest cloud VMs without specialised hardware.

Everyday Multitask Reasoning

Optimised for short reasoning chains: summarisation, classification, calendar builds, quick contract checks.

Agent Discovery Primitives

Exposes function names, argument schemas, and tool IDs in a single pass—ideal for orchestrators that must select the right sub-agent fast.

Lightweight Self-Correction

Sub-2 ms sanity loops catch arithmetic slips or JSON-schema mismatches before emitting answers.

Safety Filters Built-In

Compact rule-based guardrails screen known jailbreak strings and unsafe transaction patterns.

Typical Use-Cases

Domain	How ASI 1-Mini Excels
Customer-Support Bots	Resolves tickets, provides FAQ answers, escalates only when necessary.
Personal & Team Scheduling Agents	Parses natural-language intents, coordinates calendars, drafts reminders, and handles rescheduling conflicts.
E-commerce Concierge	Classifies products, recommends bundles, and generates short persuasive descriptions.
On-Chain Utility Helpers	Runs light contract sanity checks, surfaces obvious security smells, and suggests gas-saving refactors.
Enterprise Data Routers	Triages incoming emails or Slack messages and routes them to specialised downstream agents.
IoT Voice Assistants	Translates user commands into structured API calls for smart-home ecosystems with ultra-low latency.

Trust & Transparency

Capability	Detail
Traceable Thoughts	Returns concise reasoning snippets (≤ 4 steps) with SHA-256 hash for provenance logging.
Controlled Hallucination Rate	≤ 1.2 % unsupported claims in everyday domains after post-filtering.
Security Awareness	Flags re-entrancy or allowance-change risks when scanning ERC-20/721 snippets, recommending common-sense mitigations.

Ready to try it? Head back to the Chat Completion quick-start and select model:"asi1-mini" in your request body.