feat(providers): add Inco

[nikolas-sturm]

Summary

Adds Inco — a prepaid, OpenAI- and Anthropic-compatible inference provider (api.inco.ai).

• provider.toml: @ai-sdk/openai-compatible, INCO_API_KEY, https://api.inco.ai/v1
• logo.svg: "inco" wordmark (Montserrat Alternates SemiBold outlines) knocked out of a straight square, currentColor
• 7 model routes, all base_model + override-only (cost, reasoning options, interleaved field, display name)

model	base	input / output USD per MTok
deepseek-v4.1-flash:fast	deepseek/deepseek-v4.1-flash	0.60 / 2.40
glm-5.3	zhipuai/glm-5.3	1.40 / 4.40
glm-5.3:fast	zhipuai/glm-5.3	2.80 / 8.80
glm-5.3-flash:fast	zhipuai/glm-5.3-flash	0.15 / 0.50
kimi-k3:fast	moonshotai/kimi-k3	6.00 / 30.00
minimax-m3	minimax/MiniMax-M3	0.30 / 1.20
minimax-m3:fast	minimax/MiniMax-M3	0.60 / 2.40

:fast routes are a 2x credit premium over standard speed (documented on the Inco dashboard model list); only some models offer a standard-speed route, the rest are :fast-only.

Reasoning controls (verified against the live API)

Every response carries interleaved reasoning in reasoning_content. Probing results:

• GLM-5.3 / GLM-5.3-Flash: toggle via chat_template_kwargs.enable_thinking = true|false; reasoning on by default (works even on GLM-5.3, unlike native z.ai where it cannot be disabled)
• DeepSeek V4.1 Flash: toggle via chat_template_kwargs.thinking = true|false; reasoning is off by default on this route
• Kimi K3: toggle via chat_template_kwargs.thinking = true|false; reasoning on by default
• MiniMax M3: always reasons, no caller control — thinking.type, chat_template_kwargs.thinking/enable_thinking, and reasoning were all tested and ignored → reasoning_options = []
• reasoning_effort and thinking.type are silently ignored on all routes (invalid values are accepted without error); no effort levels are authored

Notes

• No cache pricing published ("cheaper rate applies to cached prompt tokens" per docs, no numbers), so cache_read is omitted
• No sync module — 7-model catalog, /v1/models exposes only IDs
• bun validate passes

Pricing source

Inco platform dashboard model list (https://platform.inco.ai, requires sign-in; no public pricing page exists) as reported by the contributor, cross-checked against the live API catalog.

[github-actions]

Action items

• [medium] [violation] providers/inco/models/glm-5.3:fast.toml:5 - Check: Every toggle must have a leading top-of-file comment with the exact wire path. Why: This file sets type = "toggle" but only cross-references glm-5.3.toml (“see … for controls”), so the host control surface is not documented on this entry. Action: Add a leading comment with the verified Inco wire path (e.g. # Toggle: chat_template_kwargs.enable_thinking = true|false), matching the other GLM entries.

[github-actions]

No actionable findings.