thinking/README.md

thinking

Fold how hard to think into the native shape each upstream model accepts.

Callers speak two dialects — Anthropic sends a token budget, OpenAI sends an effort ordinal — and every upstream reasons by its own control. Between them sits one neutral value, a Depth ordinal: each dialect is a constructor into it, each upstream a projection out of it.

d := thinking.Budget(32000)        // Anthropic "ultrathink" → Max
d := thinking.Effort("high")       // OpenAI reasoning_effort → High

v := thinking.Of("glm-5.2")        // GLM
d.Fields(v)                        // {"reasoning_effort": "max"}

thinking.Max.Fields(thinking.Of("kimi-k2.6"))
// {"thinking": {"type": "enabled", "preserve_thinking": "all"}}

thinking.High.Fields(thinking.Of("qwen3.5-397b-a17b"))
// {"enable_thinking": true}

One home for the map, so every serving layer folds the same way. Clear thinking.Keys on the outbound body before merging Fields, so the fold is the sole writer and no field leaks across a vocabulary boundary.

Vocabularies

Vocab	upstreams	control
GLM	GLM-5.*, DeepSeek V4, the DO-AI default	reasoning_effort ∈ {high, max}
OpenAI	o-series	reasoning_effort ∈ {low, medium, high}
Qwen	Qwen 3.*	enable_thinking gate
Kimi	Kimi K2.*	thinking object (preserve_thinking)

The fold is total and monotone — deeper in, harder out — and lossy where a vocabulary is coarser (GLM has no low/medium; OpenAI has no max).

Licensed under MIT or Apache-2.0, at your option — per HIP-0137.