SuperAPI AI Directory
Partner with us中文AI newsFavorites

AI developer platforms

3,631 results · Page 68 of 101

evalscope

AI developer platforms

EvalScope is a one-stop large model evaluation framework built by the ModelScope community, supporting model capability evaluation, inference performance stress testing, and result visualization. It comes with multiple industry evaluation benchmarks such as MMLU, C-Eval, GSM8K, etc., can evaluate various model types such as LLM, VLM, Embedding, Reranker, AIGC, and integrates evaluation backends such as OpenCompass, VLMEvalKit, RAGEval. EvalScope provides an Agent evaluation mode, inference performance testing (TTFT, TPOT and other metrics), Web Dashboard interactive reports, Arena multi-model battle mode, and supports custom datasets, models, and evaluation metric extensions.

Read more

EvoAgentX

AI developer platforms

EvoAgentX is an open-source framework for building, evaluating, and evolving LLM-based agents or agent workflows in an automated, modular, and goal-driven manner. At its core, it introduces a self-evolving agent ecosystem, enabling AI agents to be constructed, evaluated, and optimized through iterative feedback loops. Key features include: automatic construction of multi-agent workflows from a single prompt, built-in automatic evaluators, a self-evolution engine, integration of popular models such as OpenAI, Qwen, Claude, Deepseek, and kimi through LiteLLM, siliconflow, or openrouter, as well as local deployment support. The framework comes with a rich set of built-in tools (code interpreter, search, file system, database, image, browser automation, etc.), supports short-term and long-term memory modules, and supports human-in-the-loop (HITL) interaction. EvoAgentX is suitable for AI researchers, workflow engineers, and startup teams, helping to quickly build fully functional agent systems from vague ideas. The project is installed via pip and provides detailed documentation, tutorials, and examples.

Read more

EXAONE-Deep

AI developer platforms

EXAONE Deep is a series of reasoning-enhanced language models developed and released by LG AI Research, with parameter scales covering 2.4B, 7.8B, and 32B. According to the official introduction, this series performs outstandingly on reasoning benchmarks such as mathematics and programming: the 2.4B version outperforms other models of the same scale, the 7.8B version not only outperforms open-weight models of the same scale but also outperforms OpenAI o1-mini, a proprietary reasoning model, and the 32B version is competitive with leading open-weight models. These models are trained using optimized configurations, and the official recommendation is to follow the usage guide to achieve optimal performance. Users can quickly get started with Transformers, or use quantized weights in AWQ or GGUF format to run locally via llama.cpp, Ollama, or LM-Studio, or deploy with frameworks such as TensorRT-LLM, vLLM, and SGLang. The models require transformers>=4.43.1, and the latest version is recommended; when running the 32B BF16 model, it may be necessary to download and merge all shard files. The official documentation also explains the model's limitations: it may generate inappropriate, harmful, or biased content, and it does not reflect the latest information. This model uses the EXAONE AI Model License Agreement 1.1 - NC license.

Read more

experimental-ext-skills

AI developer platforms

experimental-ext-skills is the code repository for the official Model Context Protocol (MCP) Skills extension (identified as io.modelcontextprotocol/skills), maintained by the Skills Over MCP Working Group. This extension is based on the SEP-2640 proposal, with the goal of enabling skill discovery and distribution through MCP primitives. The repository also hosts the working group's working documents; the stable specification exists as published snapshots and is not edited in place directly; changes are proposed as dated entries in the decision log, and once accepted enter the draft revision of the specification and are released with a new version. The docs/ directory contains the decision log, design rationale, and implementation checklist; early design documents, findings, and tracking records are frozen in docs/archive/, none of which are part of the specification.

Read more

fastagent

AI developer platforms

FastAgent is a runtime and adapter layer that publishes a local agent directory as an online service. It does not introduce a new agent-authoring DSL, but instead reuses your existing directory structure: Markdown instructions, reusable Skills, and TypeScript tools still exist as files. FastAgent is responsible for connecting these definitions to Next/Astro applications, Telegram, Slack, Feishu/Lark, HTTP/SSE interfaces, or custom channels, providing streaming replies, group awareness, and webhook validation. It has built-in multi-model support based on the open-source pi harness (OpenAI, Anthropic, Google, etc.), tools are validated through defineTool and Zod, and Agent Skills are loaded on demand. It emphasizes a small-core philosophy: invoke decouples channels, agents, harness, and infrastructure, and does not take over your authentication, database, routing, and deployment. You can use fastagent deploy to generate containers and configurations for Docker, Fly, Railway, and AWS Bedrock AgentCore, or embed it directly into an existing application with a single route. The project is MIT licensed and is currently in a pre-1.0 version.

Read more

fastapi_mcp

AI developer platforms

FastAPI-MCP is a development library that exposes FastAPI endpoints as Model Context Protocol (MCP) tools, with built-in authentication based on FastAPI dependencies. It is implemented in a FastAPI-native way, rather than being a simple OpenAPI-to-MCP converter. With minimal configuration, it can point to a FastAPI application and run, while preserving the schemas of request models and response models as well as the Swagger documentation for all endpoints. The library supports flexible deployment: the MCP server can be mounted on the same application or deployed separately, and it uses ASGI transport to communicate efficiently directly through FastAPI's ASGI interface. Installation is recommended with uv, and pip can also be used.

Read more