Results 156 issues of OceanLi

**Issue:** **Multi-Step “Chain-of-Programs” Orchestrator** - **Description:** - Provide a multi-step orchestration pattern where each step is a distinct program snippet that refines or builds upon the previous step’s result. -...

enhancement
p1

**Issue:** **Program-of-Thought Execution & Validation** - **Description:** - Implement a specialized “program-of-thought” module where the AI agent must produce a Python snippet to validate or confirm reasoning before moving to...

enhancement
p1

director class facilitate LLM dynamic action directive composition. It render LNDL (written by LLM) translate to callable function and submit to action manager for invoke

enhancement
p1

- validator , validate output format - prompter, optimize prompts (can integrate LLMlingua here) - scorer, score output - summarizer, summarize - planner, plan and take feedback and iteratively improve...

enhancement
p1

enhancement
help wanted
p2

Though currently we use branch and session to make LLM calls and action invoke and perform simple iterative loops to apply techniques such as ReAct. The naming of these indicates...

enhancement
p0

enhancement
help wanted
p2