Apodex 1.1 Open-Source Agent Release

Apodex 1.1, an AI agent that executes complex tasks beyond deep research, was released on August 25, 2026. - It not only organizes materials and writes reports but also performs tasks through to delivery. - The open-source FrontierAgent execution framework requires no Docker and runs with a single command. - An open-weight Apodex 1.1 mini 35B model was released, with official evaluations showing competitive performance in professional, financial, and scientific benchmarks. - The company was founded and is personally led by Chen Tianqiao, with team members from Shanda Innovation Institute. - The web interface splits tasks into sub-agents with distinct roles, displays a live task board, and allows mid-execution information supplementation. - Before delivering reports, it performs 116 verifications on data, content, and sources. - The tool is suitable for building custom Deep Research products by deploying the mini model locally with FrontierAgent.
Key insights
Apodex 1.1 Goes Beyond Research to Task Execution
Unlike traditional deep research tools that only compile reports, Apodex 1.1 actively executes tasks until final delivery. It splits complex requests into subtasks, creates role-based sub-agents, and shows real-time progress on a task board. This execution-oriented design makes it suitable for workflows that require not just information synthesis but also action completion.
Open-Source Model and Framework Enable Custom Agent Builds
The release includes the FrontierAgent execution framework (no Docker, one-command run) and the Apodex 1.1 mini 35B model weights. Developers can locally deploy the model with the framework to create their own deep research or task-execution agents. This lowers the barrier for building specialized AI products without relying on cloud APIs.
Apodex Is Backed by Chen Tianqiao and Shanda Innovation Institute Talent
The company behind Apodex was founded and is personally led by Chen Tianqiao, the founder of Shanda. Many team members are alumni of the Shanda Innovation Institute, often referred to as the 'Whampoa Military Academy' of China's internet industry. This pedigree suggests strong R&D capability and long-term commitment to agentic AI.
Transparent Multi-Agent Execution with 116-Verification Step Builds Trust
During task execution, users can see which sub-agent is working on which task and can supplement requirements in real time. Before delivering the final report, the system performs 116 checks on data accuracy, content quality, and source reliability. This transparency and verification process addresses common trust issues in AI-generated outputs.
DeepSeek Harness Integration Evaluation Suggests Waiting for Maturity
A user tested Apodex 1.1 while evaluating whether to integrate DeepSeek Harness into their own product. The agent's analysis concluded that DeepSeek Harness is iterating very quickly, and formal integration should wait, though individuals can experiment with it. This demonstrates Apodex's capability to provide actionable technical assessments.