Open Models Almanac
Methodsinference & servingagentic scaffolding

single mention · inference & serving

Cold-start prompting for reasoning in tool use

Cold-start prompting integrates a model's reasoning and tool-use capabilities by designing system prompts that instruct it to interleave tool calls within its reasoning process.1,2 It relies on the model following explicit instructions, so it occasionally produces the desired reasoning-in-tool-use trajectories that can serve as a basis for later training.3,4

Trade-off The approach works even though the resulting reasoning-in-tool-use patterns may lack robustness.5

How this definition was written

Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.

  1. a straight- forward strategy for integrating these two capabilities is through carefully designed prompting — paper
  2. the system prompt we designed to instruct the model to incorporate multiple tool calls within its reasoning process — paper
  3. We posit that the model possesses sufficient ability to accurately follow explicit instructions — paper
  4. the model is occasionally able to generate the desired trajectories, thereby providing a basis for subsequent — paper
  5. although the reasoning in tool-use patterns may lack robustness — paper

Also called Cold-Start.

Filed under inference & serving › agentic scaffolding