SuperAPI AI Directory
Partner with us中文AI newsFavorites

AI developer platforms

3,662 results · Page 38 of 102

recurrentgemma

AI developer platforms

RecurrentGemma is a series of open-weight language models released by Google DeepMind, based on the novel Griffin architecture. This architecture replaces global attention with a hybrid mechanism of local attention and linear recurrence, thereby achieving fast inference when generating long sequences. This repository contains the model implementation as well as example code for sampling and fine-tuning. The official recommendation is for most users to use the highly optimized Flax implementation, while an unoptimized PyTorch reference implementation is also provided. Model checkpoints can be downloaded via Kaggle and include weights and tokenizer. The code supports running on CPU, GPU, or TPU; the Flax implementation is optimized for TPU and includes low-level Pallas kernels for linear scans of recurrent layers. The project uses Poetry to manage dependencies and provides Colab tutorials covering sampling and fine-tuning.

Read more

Reflexio

AI developer platforms

Reflexio is a learning platform for AI agents, designed to let agents learn from user corrections, failed paths, and successful outcomes, so they improve their behavior over time and avoid repeating mistakes. It is not a traditional memory layer, but a behavior learning platform: agents publish actual interaction processes, Reflexio extracts reusable behavior rules, and these rules are read back on the next run, all without retraining the model. The platform supports wrapping existing LLM calls through a lightweight SDK, and can also be integrated via Python, REST, or CLI; the extracted signals include trigger conditions and feedback, and can be precisely injected into the relevant context to control token costs. Every learning can be viewed and reviewed, and can be approved, set to pending, or rejected, with rejected ones removed from retrieval. Official deployment options range from fully managed, bring-your-own model credentials, to BYOC/self-hosted, and learning data can be stored in the user's own database.

Read more

Reka AI

AI developer platforms

Reka AI is an AI research lab that builds native multimodal models (Spark, Edge, Flash, Core) capable of processing video, images, audio, and text. Its models are used by enterprises in the security, media, and defense sectors. Reka AI is committed to building models and infrastructure for the era of physical AI; its Omni model can natively process and generate various types of data within a single architecture and runs in real time. The company also provides scalable infrastructure for annotating, inferring on, searching, and clipping large-scale video data, accessible via API, MCP, or applications. In addition, Reka AI provides the training data behind frontier AI, including first-person video, robot trajectories, world model materials, and large-scale expert human judgment.

Read more

RelayPulse

AI developer platforms

RelayPulse is an LLM relay service monitoring platform for developers, officially positioned to monitor in real time the availability, latency, and sponsorship links of global LLM relay services. By continuously detecting interface connectivity, it helps developers quickly assess the service quality of different providers and thereby discover more stable API providers. According to the official introduction, the platform supports connectivity monitoring for mainstream model interfaces such as Claude, GPT, and Gemini, primarily addressing the problem that developers find it difficult to judge service stability when using third-party relay APIs. Currently, the available information does not provide details such as specific monitoring metrics, data refresh frequency, whether it is paid, the number of supported relay providers, whether registration or login is required, or specific usage steps; this content is not yet available.

Read more

Remote-MCP

AI developer platforms

Remote-MCP is a type-safe solution for remote MCP (Model Context Protocol) communication, developed by ssut and open-sourced on GitHub. It provides bidirectional, concise remote communication capabilities, allowing local MCP clients to connect to remote MCP servers, thereby enabling remote access to and centralized management of model context. The project includes two core packages, @remote-mcp/client and @remote-mcp/server: the client library can run as a local MCP server and connect to a remote implementation, while the server library is used to create remotely accessible MCP services. It supports basic MCP commands, tools, and Prompt capabilities, and provides mechanisms such as custom HTTP headers, authentication middleware, basic middleware, and basic error handling, while also planning to support a full event subscription system (resource change, tool/prompt list change notifications), crash-safe handling, and more. The project also provides implementation examples such as Cloudflare Workers and standalone Node.js, and includes a Nest.js integration package. This library is a supplementary extension to the official MCP specification and is not part of the official specification. It is licensed under the MIT License. The project is currently in active development and is experimental, and breaking changes may occur.

Read more

RestGPT

AI developer platforms

RestGPT is an autonomous agent based on large language models, designed to control real-world applications such as movie databases and music players through RESTful APIs. The project was released by Yifan-Song793 on GitHub and includes the paper's code. RestGPT adopts an iterative coarse-to-fine online planning framework and introduces an executor to invoke RESTful APIs. Its core components include a planner (which generates a natural language subtask for the current step), an API selector (which maps coarse-grained high-level subtasks into finer-grained API call plans), and an executor (which executes the API call plan, where the caller organizes parameters according to the API plan and documentation, and the parser generates Python code according to the response schema to parse the response). To evaluate performance, the project also proposes the RestBench benchmark, which includes two real-world scenarios, the TMDB movie database and the Spotify music player, as well as manually annotated instructions and golden solution paths. The code has been released, and the demo is under construction.

Read more