Crof-AI-OpenCode

Working reference for using Crof.ai as a safe OpenCode custom provider.

This repo documents a configuration that was tested against Crof's OpenAI-compatible API and shaped to work well with OpenCode without relying on undocumented behavior.

It also includes a reusable opencode.json template with the Crof provider only.

Quick Start

Most users will want to copy the provider config directly into their own opencode.json.

Use this template as-is, then add your Crof API key separately in OpenCode auth storage at ~/.local/share/opencode/auth.json.

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "crof": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "CrofAI",
      "options": {
        "baseURL": "https://crof.ai/v1"
      },
      "models": {
        "moonshotai/kimi-k2.5": {
          "id": "kimi-k2.5",
          "name": "Kimi K2.5",
          "limit": { "context": 262144, "output": 262144 },
          "family": "kimi",
          "attachment": true,
          "reasoning": true,
          "tool_call": true,
          "temperature": true,
          "modalities": { "input": ["text", "image", "video"], "output": ["text"] },
          "interleaved": { "field": "reasoning_details" }
        },
        "moonshotai/kimi-k2.5:lightning": {
          "id": "kimi-k2.5-lightning",
          "name": "Kimi K2.5 (lightning)",
          "limit": { "context": 131072, "output": 32768 }
        },
        "z-ai/glm-5": {
          "id": "glm-5",
          "name": "GLM-5",
          "limit": { "context": 202752, "output": 202752 },
          "family": "glm",
          "attachment": false,
          "reasoning": true,
          "tool_call": true,
          "temperature": true,
          "modalities": { "input": ["text"], "output": ["text"] },
          "interleaved": { "field": "reasoning_content" }
        },
        "z-ai/glm-5:lightning": {
          "id": "glm-5-lightning",
          "name": "GLM-5 (lightning)",
          "limit": { "context": 202752, "output": 202752 }
        },
        "z-ai/glm-4.7": {
          "id": "glm-4.7",
          "name": "GLM-4.7",
          "limit": { "context": 202752, "output": 202752 },
          "family": "glm",
          "attachment": false,
          "reasoning": true,
          "tool_call": true,
          "temperature": true,
          "modalities": { "input": ["text"], "output": ["text"] },
          "interleaved": { "field": "reasoning_details" }
        },
        "z-ai/glm-4.7-flash": {
          "id": "glm-4.7-flash",
          "name": "GLM-4.7-Flash",
          "limit": { "context": 202752, "output": 131072 },
          "family": "glm",
          "attachment": false,
          "reasoning": true,
          "tool_call": true,
          "temperature": true,
          "modalities": { "input": ["text"], "output": ["text"] },
          "interleaved": { "field": "reasoning_details" }
        },
        "google/gemma-4-31b-it": {
          "id": "gemma-4-31b-it",
          "name": "Gemma 4 31B",
          "limit": { "context": 262144, "output": 262144 },
          "family": "gemma",
          "attachment": true,
          "reasoning": true,
          "tool_call": true,
          "temperature": true,
          "modalities": { "input": ["text", "image", "video"], "output": ["text"] }
        },
        "minimax/minimax-m2.5": {
          "id": "minimax-m2.5",
          "name": "MiniMax M2.5",
          "limit": { "context": 204800, "output": 131072 },
          "family": "minimax",
          "attachment": false,
          "reasoning": true,
          "tool_call": true,
          "temperature": true,
          "modalities": { "input": ["text"], "output": ["text"] },
          "interleaved": { "field": "reasoning_details" }
        },
        "qwen/qwen3.5-397b-a17b": {
          "id": "qwen3.5-397b-a17b",
          "name": "Qwen3.5 397B A17B",
          "limit": { "context": 262144, "output": 262144 },
          "family": "qwen",
          "attachment": true,
          "reasoning": true,
          "tool_call": true,
          "temperature": true,
          "modalities": { "input": ["text", "image", "video"], "output": ["text"] }
        },
        "deepseek/deepseek-v3.2": {
          "id": "deepseek-v3.2",
          "name": "DeepSeek V3.2",
          "limit": { "context": 163840, "output": 163840 },
          "family": "deepseek",
          "attachment": false,
          "reasoning": true,
          "tool_call": true,
          "temperature": true,
          "modalities": { "input": ["text"], "output": ["text"] }
        }
      }
    }
  }
}

Goal

The goal of this setup is to give OpenCode users a Crof provider that is:

compatible with OpenCode's custom provider system
aligned with OpenRouter-style model naming where possible
sourced from Crof's own documented model data for token limits
conservative about metadata so it stays safe and predictable
cleaned up to exclude models that did not work reliably in OpenCode

Source Of Truth

This setup was built from two sources:

https://crof.ai/docs
https://models.dev/api.json

Each source was used for a different purpose.

Crof Docs

Crof's docs were used as the source of truth for:

base URL
OpenAI-compatible API behavior
supported request parameters
reasoning support via reasoning_effort
model discovery through /v1/models
token limits through context_length and max_completion_tokens

From Crof's docs, the provider is configured against:

https://crof.ai/v1

And model limits are taken from the documented /v1/models response shape.

Models.dev

models.dev was used to make the provider feel more natural inside OpenCode.

It was used for:

canonical OpenRouter-style model keys where exact matches exist
matching display names to known model naming conventions
safe manual metadata for exact-match models

That metadata includes:

family
attachment
reasoning
tool_call
temperature
modalities
interleaved when available

This was only applied where there was an exact canonical match. The Lightning variants (Kimi K2.5 Lightning and GLM-5 Lightning) are intentionally downgraded: they keep tool_call: true and temperature: true but drop reasoning, attachment, interleaved, and multimodal input because they are speed-optimized models that trade off reasoning depth and vision capabilities for faster responses.

Models API Verification

Crof's /v1/models endpoint returns the following fields for each model:

id
context_length
max_completion_tokens
pricing

The /v1/models response does not include capability metadata like reasoning, tool_call, modalities, or attachment. Those fields must be inferred and set manually in the OpenCode config.

All limit.context and limit.output values in this config were verified against the live /v1/models API response. The following table shows the exact match:

Model	API `context_length`	API `max_completion_tokens`	Config `limit.context`	Config `limit.output`
`kimi-k2.5`	262144	262144	262144	262144
`kimi-k2.5-lightning`	131072	32768	131072	32768
`glm-5`	202752	202752	202752	202752
`glm-5-lightning`	202752	202752	202752	202752
`glm-4.7`	202752	202752	202752	202752
`glm-4.7-flash`	202752	131072	202752	131072
`gemma-4-31b-it`	262144	262144	262144	262144
`minimax-m2.5`	204800	131072	204800	131072
`qwen3.5-397b-a17b`	262144	262144	262144	262144
`deepseek-v3.2`	163840	163840	163840	163840

Lightning Model Limitations

The Lightning variants are speed-optimized models. They differ from their base counterparts in important ways:

No reasoning: Lightning models do not produce extended reasoning output. They are configured with reasoning: false and no interleaved field.
No vision: Lightning models are text-only. They are configured with attachment: false and modalities.input: ["text"].
Lower output limits: Kimi K2.5 Lightning has a much smaller output limit (32768 vs 262144) which can cause truncation on long responses.

If you experience pausing or failed tool calls with Lightning models, switch to the base model (e.g. moonshotai/kimi-k2.5 instead of moonshotai/kimi-k2.5:lightning).

Why Lightning Models Were Downgraded

The Lightning entries were initially configured with the same reasoning: true, attachment, interleaved, and multimodal settings as their base models. This caused two problems in practice:

Model pausing: OpenCode expected reasoning output (via interleaved) that Lightning models do not produce, causing the stream to hang while waiting for reasoning tokens that never arrive.
Failed tool calls: Lightning models struggle with complex tool-calling formats. Marking them as reasoning: true made OpenCode expect deeper agentic behavior that the speed-optimized models cannot reliably deliver.

The fix was to downgrade Lightning metadata to match reality:

reasoning: false — no extended thinking output
attachment: false — no image or video input
interleaved removed — no reasoning stream field
modalities.input: ["text"] — text-only

This prevents OpenCode from expecting capabilities the Lightning models do not have, eliminating the pausing and reducing tool call failures.

Why This Is A Safe OpenCode Custom Provider

OpenCode supports custom providers through provider.<id> entries in opencode.json using @ai-sdk/openai-compatible.

For this setup, the provider is:

provider ID: crof
adapter: @ai-sdk/openai-compatible
base URL: https://crof.ai/v1

The config uses a conservative approach:

use Crof as the source of truth for API compatibility and limits
use models.dev only where there is a clear exact match
use id overrides when the Crof backend model name differs from the displayed OpenCode model key
avoid inventing metadata for Crof-only variants
remove models that failed in practice with OpenCode

This means the configuration aims to be both practical and low-risk.

Files Used By OpenCode

Global provider config:

~/.config/opencode/opencode.json

Stored provider credentials:

~/.local/share/opencode/auth.json

Template included in this repo:

./opencode.json

The repo template contains no API key.

Working Models In This Setup

Exact OpenRouter-style matches:

moonshotai/kimi-k2.5 -> Kimi K2.5
z-ai/glm-5 -> GLM-5
z-ai/glm-4.7 -> GLM-4.7
z-ai/glm-4.7-flash -> GLM-4.7-Flash
google/gemma-4-31b-it -> Gemma 4 31B
minimax/minimax-m2.5 -> MiniMax M2.5
qwen/qwen3.5-397b-a17b -> Qwen3.5 397B A17B
deepseek/deepseek-v3.2 -> DeepSeek V3.2

Crof-only variants retained through id overrides:

moonshotai/kimi-k2.5:lightning -> id: kimi-k2.5-lightning
z-ai/glm-5:lightning -> id: glm-5-lightning

Display names are harmonized so the list looks consistent in OpenCode:

Kimi K2.5
Kimi K2.5 (lightning)
GLM-5
GLM-5 (lightning)
GLM-4.7
GLM-4.7-Flash
Gemma 4 31B
MiniMax M2.5
Qwen3.5 397B A17B
DeepSeek V3.2

Models Removed

These models were intentionally removed from the final config:

canopy variants: removed because they were not working reliably in OpenCode
stok-0.4.1: removed because it did not work correctly with OpenCode

The repo and template document only the remaining working set.

Reasoning And Thinking Support

Crof's docs explicitly document support for:

reasoning model output
reasoning streaming fields
reasoning_effort

Accepted Crof reasoning effort values:

low
medium
high
none

Important detail:

none disables the reasoning phase entirely
higher values allow more thinking before the final answer

In this setup, exact-match reasoning-capable models are marked with:

reasoning: true

And models with known reasoning stream field behavior also include interleaved where models.dev provides it.

Examples:

moonshotai/kimi-k2.5 -> interleaved.field = reasoning_details
z-ai/glm-5 -> interleaved.field = reasoning_content
z-ai/glm-4.7 -> interleaved.field = reasoning_details
z-ai/glm-4.7-flash -> interleaved.field = reasoning_details
minimax/minimax-m2.5 -> interleaved.field = reasoning_details

The Lightning variants (Kimi K2.5 Lightning and GLM-5 Lightning) are intentionally configured with reasoning: false, no interleaved field, and text-only modalities because they are speed-optimized models.

Limits

Crof's docs describe /v1/models as returning per-model:

context_length
max_completion_tokens

Those values are used directly for:

limit.context
limit.output

At the time this config was verified, the configured limits matched the live Crof /v1/models response for the working models.

Why The `id` Field Matters

OpenCode uses the key in provider.<provider>.models as the selected model identifier.

For custom providers, id lets you send a different upstream model name to the provider API.

Example:

{
  "moonshotai/kimi-k2.5:lightning": {
    "id": "kimi-k2.5-lightning",
    "name": "Kimi K2.5 (lightning)"
  }
}

This keeps the OpenCode-facing key clean and consistent while still sending the Crof backend model ID that actually works.

What Was Added Manually

For exact-match models, this setup adds safe manual metadata using fields supported by the OpenCode config schema:

family
attachment
reasoning
tool_call
temperature
modalities
interleaved

These values were copied from models.dev only when there was an exact model match.

What Was Not Added

Some things were intentionally not added or not assumed.

no API keys are stored in this repo
no default model is forced in the template
no metadata was guessed for Crof-only variants beyond id, name, and limit, except for Lightning variants which are intentionally downgraded to speed-optimized settings (no reasoning, no vision, text-only)
structured_output was not added because it does not appear to be supported on custom provider model entries in the current OpenCode schema

How To Use This Globally

Merge the provider.crof block from this repo's opencode.json into ~/.config/opencode/opencode.json.
Add a crof credential entry to ~/.local/share/opencode/auth.json.
Restart OpenCode or reopen the session.
Select a model under the crof provider.

Example auth entry:

{
  "crof": {
    "type": "api",
    "key": "YOUR_CROF_API_KEY"
  }
}

How To Use This Per Project

If you want project-level config instead of global config, use this repo's opencode.json as your starting point.

You still need a valid crof credential in OpenCode auth storage unless you provide credentials some other way.

Verification Performed

This setup was checked using the following process:

Crof docs were reviewed for API behavior, supported parameters, and /v1/models response structure.
OpenCode provider docs and config schema were reviewed for custom provider support and model field compatibility.
The live Crof /v1/models endpoint was queried to verify model IDs and limits.
A live request to https://crof.ai/v1/chat/completions succeeded with a working model.
Broken models were removed after direct testing in OpenCode.

Example verification request:

curl -sS \
  -H "Authorization: Bearer YOUR_CROF_API_KEY" \
  -H "Content-Type: application/json" \
  "https://crof.ai/v1/chat/completions" \
  -d '{
    "model": "deepseek-v3.2",
    "messages": [{"role": "user", "content": "Reply with exactly: ok"}],
    "max_tokens": 8,
    "temperature": 0
  }'

Security

this repo contains no real API key
credentials belong in ~/.local/share/opencode/auth.json, not in the repo
the included opencode.json template is safe to share

Included File

opencode.json: Crof-only OpenCode custom provider template

Name		Name	Last commit message	Last commit date
Latest commit History 9 Commits
.github/workflows		.github/workflows
.vale/styles/Crof		.vale/styles/Crof
scripts		scripts
.gitignore		.gitignore
.markdownlint.json		.markdownlint.json
.vale.ini		.vale.ini
README.md		README.md
opencode.json		opencode.json
package.json		package.json

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

Crof-AI-OpenCode

Quick Start

Goal

Source Of Truth

Crof Docs

Models.dev

Models API Verification

Lightning Model Limitations

Why Lightning Models Were Downgraded

Why This Is A Safe OpenCode Custom Provider

Files Used By OpenCode

Working Models In This Setup

Models Removed

Reasoning And Thinking Support

Limits

Why The `id` Field Matters

What Was Added Manually

What Was Not Added

How To Use This Globally

How To Use This Per Project

Verification Performed

Security

Included File

About

Uh oh!

Releases

Packages

Uh oh!

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

Crof-AI-OpenCode

Quick Start

Goal

Source Of Truth

Crof Docs

Models.dev

Models API Verification

Lightning Model Limitations

Why Lightning Models Were Downgraded

Why This Is A Safe OpenCode Custom Provider

Files Used By OpenCode

Working Models In This Setup

Models Removed

Reasoning And Thinking Support

Limits

Why The id Field Matters

What Was Added Manually

What Was Not Added

How To Use This Globally

How To Use This Per Project

Verification Performed

Security

Included File

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Uh oh!

Contributors

Uh oh!

Languages

Why The `id` Field Matters

Packages