SuperAPI AI Directory
Partner with us中文AI newsFavorites

All AI tools

35,702 results · Page 959 of 992

Mini SWE-Agent

AI agents

Mini SWE-Agent is an open-source AI coding agent built by the SWE-bench and SWE-agent teams (Princeton and Stanford), implemented in only about 100 lines of Python code, positioned as a minimal agent that can be widely adopted, understood, and extended. It does not rely on complex tools and dependency libraries, performing all operations solely through bash, and does not use the language model's tool-calling interface, so it can run with almost any model. Officially, it is said to score over 74% on the SWE-bench verified benchmark and to start far faster than Claude Code. The project has been adopted by Meta, NVIDIA, IBM, Stanford University, Princeton University, and others, supports multiple deployment methods such as local environments, docker/podman, singularity/apptainer, and bubblewrap, and can be compatible with multiple models through litellm, openrouter, portkey, etc., supporting the /completion and /response endpoints as well as interleaved thinking. Its history is completely linear, and actions are executed independently via subprocess.run, making it easy to debug, fine-tune, extend, and benchmark, and suitable as a baseline system or an agent framework that emphasizes the capabilities of the language model itself.

Read more

MINI_LLM

AI developer platforms

MINI_LLM is a personal practice project for experimenting with and reproducing the pretraining pipeline of a small-parameter Chinese large language model. The project's base model is QWEN, and a pretrained model with approximately 1.4B parameters was trained, with about 8B training tokens. The repository covers the complete pipeline of pretraining, SFT instruction fine-tuning, and DPO (Direct Preference Optimization), and includes DeepSpeed distributed training techniques; the PPO part is marked as to be done. The project provides download links for model weights at each stage of pretraining, SFT, and DPO, and uses SwanLab to view training logs. It is suitable for developers who want to learn or reproduce the Chinese large model training pipeline.

Read more

MiniLLaMA3

AI developer platforms

MiniLLaMA3 is a mini version project of LLaMA3, aimed at providing a complete training pipeline for a large language model from zero to one. The project covers stages such as constructing data, training the tokenizer, pretraining (PT), supervised fine-tuning (SFT), and direct preference optimization (DPO), helping developers understand and practice the entire process of large model training. The project is hosted on GitHub and is suitable for learning and research purposes.

Read more

Mintli

AI marketing

Mintli is an AI ad and marketing content generator for UK small businesses. Users simply paste their website address, and the system reads the brand colors, logo, tone of voice, and the products actually sold on the website, then generates brand-matched ad images, ad copy, email marketing campaigns, and short videos accordingly. The tool requires no pre-assembled brand kit and no starting from templates. The copy is built on seven proven copywriting frameworks such as AIDA, PAS, and BAB, and outputs British English by default, with pricing in pounds sterling: the free tier offers 25 credits with no bank card required, followed by Starter at £9 per month, Growth at £29 per month, and Pro at £79 per month. Key features include AI copywriting, AI image generation customized to brand colors and style, email marketing campaigns with professional templates, and multichannel ad creation for Facebook, Instagram, Google, and LinkedIn. It is suitable for users who need professional marketing but do not want to pay long-term agency fees or lack design skills, such as construction and renovation, clinics, salons, agencies, retailers, and service businesses.

Read more

mirrord

Other AI tools

mirrord is a Kubernetes-oriented development platform that lets developers and AI agents (such as Claude Code, Cursor, and Copilot) connect directly to live APIs, databases, and services in a real staging cluster for testing while writing code, without deploying or waiting for CI, thereby speeding up the development loop by 10x. It manages sessions, RBAC, traffic routing, and isolation through a Helm-installed Operator, supports any language, IDE, and Kubernetes cluster, can start in 15 seconds, and is free to get started.

Read more

Miscellaneous AI Tools: Diverse Solutions for Your Unique Challenges

AI marketing

Giftpack is an AI tool for marketing, sales, HR, and customer teams, positioned as driving user retention, loyalty, and growth at scale through AI. The official website text shows that companies of all sizes use Giftpack to build relationships, automate operations, and increase retention and recurring revenue. The materials do not provide specific feature details, pricing model, price, platform support, or usage steps.

Read more

Mistral AI Agent

AI agents

Mistral AI Agent is the agent (AI Agent) building and usage capability provided by Mistral AI, and is part of its Agents and Conversations API. According to the official documentation, an AI agent is an autonomous system driven by a large language model that, after receiving high-level instructions, can plan on its own, call tools, execute processing steps, and take actions to achieve specific goals, and can also collaborate with one another to complete more complex tasks. This API enables developers to build such agents and provides multimodal models (text and vision models), persistent state across sessions, the ability to converse with foundation models/single agents/multiple agents, built-in connector tools (code execution, web search, image generation, document library), and handoff capability between agents. Features supported by the chat completion endpoint also apply, such as structured output, document understanding, tool calling, and citations.

Read more

MM-EUREKA

AI developer platforms

MM-EUREKA is an open-source multimodal reasoning research project by ModalMinds that explores the frontiers of multimodal reasoning through rule-based reinforcement learning (RL). The project has released two multimodal reasoning models, MM-Eureka-Qwen-7B and MM-Eureka-Qwen-32B, and has open-sourced the complete training pipeline, code, data, and more. On the MMK12 evaluation set, the 7B model scores 66.1 and the 32B model scores 72.3; on MathVista, they reach 73.0 and 74.8 respectively, surpassing open-source models of the same scale and larger on some metrics. The project also proposes the CPGD reinforcement learning algorithm and the MM-PRM process reward model, and has publicly released the MMK12 dataset containing 15,000 multidisciplinary K12 samples and 2,000 multiple-choice questions for reasoning evaluation in subjects such as mathematics, physics, chemistry, and biology.

Read more

mmaGPT

AI video

mmaGPT is an AI coach analysis tool for combat training videos. Users upload 20- to 60-second phone training clips (such as hitting a heavy bag, focus mitts, or sparring), and receive a technique correction report sorted by priority, including the time points where the issues occur (such as 0:02, 0:09), specific descriptions, and a personal practice drill for the next training session. It supports multiple combat styles, can be used in a browser with no download required, and covers phone, tablet, and desktop. The company says new accounts can get one free analysis with no credit card required; paid plans are priced in USD, with annual pricing including tiers of $15.83, $24.17, and $40.83 per month, as well as a Lite plan at $9 per month including 3 analyses.

Read more