avlcode/avlcode

AVL Code is an AI powered desktop assistant for coding and security analysis, made by the Landi team at Antiy, a Chinese security company. It runs natively on Windows, macOS, and Linux and combines a conversational AI assistant with a real engineer's workbench: it can read and write code, run commands, and help triage malware samples. Approval requests and task progress can follow you to your phone through WeChat, Feishu, or DingTalk. It is powered by Landi's own threat detection model along with Antiy's AVL security engine. Important context: this specific repository does not contain the product's source code. It exists only to publish release notes and to track community issues and bug reports, so there is nothing here to clone and build. All release artifacts are covered by a proprietary license and a separate end user license agreement, and the README explicitly states that no open source license is granted. The tool is described as being in the same category as Claude Code and OpenAI Codex, positioned as a domestically developed, self-controlled alternative for users in China. It is designed to be compatible with existing Claude Code style conventions, automatically picking up AGENTS.md and CLAUDE.md project instruction files and supporting the same kind of subagent naming, so existing workflows can carry over. On the security side, it includes a read only toolset for analyzing binaries, covering hashing and entropy checks, extracting indicators of compromise, parsing PE, ELF, and Mach-O file formats, disassembly, decompiling C and Java pseudocode, and matching YARA rules, any tool capable of executing files is force disabled inside its sample analysis area. Security features include local credentials encrypted with a Chinese national cryptography standard, signed skill packages, and sessions that are stored entirely on the local machine, with an enterprise edition offering full private deployment. The assistant supports five working modes, described as automatic, plan, prepare, execute, and evaluate, intended for longer running tasks beyond just writing code. It can connect to Landi's own models through a dedicated protocol, or to other models using OpenAI or Anthropic compatible protocols. The product is currently in alpha testing, and the README warns to only download installers from official channels to avoid tampered copies.