[KongchangAI]
Product

DeepFlow

An open-source Agent Harness project built on LangGraph. It implements the standard Model-Tools loop and inserts extension points into the loop through a Middleware mechanism, supporting capabilities such as context compression.

Timeline (last 90 days)

Oct 31

DeepFlow implements the standard Model-Tools loop based on LangGraph and inserts extension points into the loop through a Middleware mechanism.

Unverified50%
Oct 31

DeepFlow's Before Agent Middleware organizes static content such as the current date and System Prompt before the Agent starts, ensuring the Prefix Context stays fixed and thereby saving Tokens.

Unverified50%
Oct 31

DeepFlow offers two modes, Docker and Local Sandbox; in Local mode path virtualization is implemented, but the model can bypass the virtual paths through code concatenation, so DeepFlow enables Docker mode by default, while Local Sandbox must be actively enabled and has Bash disabled by default.

Unverified50%
Oct 31

DeepFlow's Memory mechanism is an After Agent Middleware that, after each round of dialogue, goes through three filtering steps: keeping only the user's initial input and the AI's final output, filtering out pleasantry rounds, and detecting user calibration or preference signals, after which an independent model is called to judge the persistence of the information and output a confidence score, and only when the threshold is met is it written to Agent.md.

Unverified50%

All Facts (4)

Source Articles