More coding tools: ModelFold setup

Higher credit usage on the first session is normal

Agents attach fixed tool definitions and a system prompt to every request, often tens of thousands of tokens. On the first session this content is written to the cache and billed as full input, so credit usage is higher. Later requests hit the cache and are billed at the much lower cached rate, so usage drops noticeably. This is expected behavior — you are not being overcharged. For high-volume usage, please contact support — we can offer discounts.

These tools have documented custom endpoints but are not individually featured in the main directory. Configuration below is based on official docs, not a client test. Install from the linked official project, configure a ModelFold API key and the exact model ID, then send one short OK prompt and check your account usage. For CLI snippets, set MODELFOLD_API_KEY first. Keep literal keys out of version control.

Command Code

Setup location: ~/.commandcode/providers.json. Verification: Configured from official documentation; not tested in this client.

Install the official CLI, select the provider via /connect, and switch models.

{
  "provider": {
    "modelfold": {
      "baseURL": "https://modelfold.ai/anthropic/v1",
      "apiKey": "$MODELFOLD_API_KEY",
      "api": "anthropic-messages",
      "models": {
        "claude-opus-5-5": {}
      }
    }
  }
}

For Claude, prefer the native Anthropic route. Cache savings require eligible repeated prefixes and client cache support; a custom endpoint alone does not guarantee a cache hit.

Official configuration reference.

Maki

Setup location: ~/.config/maki/providers.toml. Verification: Configured from official documentation; not tested in this client.

Install Maki and add a distinct provider slug; select modelfold/claude-opus-5-5.

[modelfold]
protocol = "anthropic"
base_url = "https://modelfold.ai/anthropic"
api_key_env = "MODELFOLD_API_KEY"
[[modelfold.models]]
id = "claude-opus-5-5"
context_window = 200000
max_output_tokens = 8192

For Claude, prefer the native Anthropic route. Cache savings require eligible repeated prefixes and client cache support; a custom endpoint alone does not guarantee a cache hit.

Official configuration reference.

DeepSeek Reasonix

Setup location: Reasonix config.toml. Verification: Configured from official documentation; not tested in this client.

Install Reasonix and add a provider; supply credentials via the referenced environment variable.

[[providers]]
name = "modelfold"
kind = "anthropic"
base_url = "https://modelfold.ai/anthropic"
models = ["claude-opus-5-5"]
default = "claude-opus-5-5"
api_key_env = "MODELFOLD_API_KEY"

For Claude, prefer the native Anthropic route. Cache savings require eligible repeated prefixes and client cache support; a custom endpoint alone does not guarantee a cache hit.

Official configuration reference.

Tabby

Setup location: ~/.tabby/config.toml. Verification: Configured from official documentation; not tested in this client.

Install the self-hosted server and editor extension. Configure the chat HTTP connector and check its System status.

[model.chat.http]
kind = "openai/chat"
model_name = "gpt-6-1-sol"
api_endpoint = "https://modelfold.ai/v1"
api_key = "YOUR_MODELFOLD_API_KEY"

Chat, completion and embeddings are separate; this config covers chat only.

Claude conversations through the OpenAI-compatible route do not use the native Claude prompt cache. For long tasks, consider a model with automatic caching or a client with native Anthropic support.

Official configuration reference.

Twinny

Setup location: Providers → OpenAI-compatible server. Verification: Configured from official documentation; not tested in this client.

Install Twinny and add an OpenAI-compatible server: HTTPS, host modelfold.ai, port 443, path /v1, model ID and key.

Provider: OpenAI Compatible / Chat Completions
URL: https://modelfold.ai/v1
API Key: YOUR_MODELFOLD_API_KEY
Model ID: gpt-6-1-sol

Use the custom server with editable host fields, not a fixed hosted preset. Configure Chat only.

Claude conversations through the OpenAI-compatible route do not use the native Claude prompt cache. For long tasks, consider a model with automatic caching or a client with native Anthropic support.

Official configuration reference.

gptme

Setup location: Environment. Verification: Configured from official documentation; not tested in this client.

Install gptme and use the local/ model prefix with your custom endpoint.

export OPENAI_BASE_URL=https://modelfold.ai/v1
export OPENAI_API_KEY="$MODELFOLD_API_KEY"
gptme -m local/gpt-6-1-sol

For Claude, prefer the native Anthropic route. Cache savings require eligible repeated prefixes and client cache support; a custom endpoint alone does not guarantee a cache hit.

Official configuration reference.

Open Interpreter

Setup location: interpreter --api_base. Verification: Configured from official documentation; not tested in this client.

Install open-interpreter using the official Python instructions. Set --api_base and --model; retain code execution confirmation.

export OPENAI_API_KEY="$MODELFOLD_API_KEY"
interpreter --model openai/gpt-6-1-sol --api_base https://modelfold.ai/v1

Claude conversations through the OpenAI-compatible route do not use the native Claude prompt cache. For long tasks, consider a model with automatic caching or a client with native Anthropic support.

Official configuration reference.

Theia AI / Theia IDE

Setup location: Settings → AI Features → OpenAI Custom. Verification: Configured from official documentation; not tested in this client.

Install Theia IDE, enable AI features, add model/url/apiKey under custom providers, and assign the model to an agent.

{
  "ai-features.openAiCustom.customOpenAiModels": [
    {
      "model": "gpt-6-1-sol",
      "url": "https://modelfold.ai/v1",
      "apiKey": "YOUR_MODELFOLD_API_KEY",
      "id": "modelfold",
      "developerMessageSettings": "system"
    }
  ]
}

For Claude, prefer the native Anthropic route. Cache savings require eligible repeated prefixes and client cache support; a custom endpoint alone does not guarantee a cache hit.

Official configuration reference.

PearAI

Setup location: PearAI model configuration → apiBase. Verification: Configured from official documentation; not tested in this client.

Install PearAI and use the OpenAI adapter with apiBase, apiKey and model.

Provider: OpenAI Compatible / Chat Completions
URL: https://modelfold.ai/v1
API Key: YOUR_MODELFOLD_API_KEY
Model ID: gpt-6-1-sol

Official source confirms apiBase overrides. Released builds were not tested; check repository maintenance status.

Claude conversations through the OpenAI-compatible route do not use the native Claude prompt cache. For long tasks, consider a model with automatic caching or a client with native Anthropic support.

Official configuration reference.

ForgeCode

Setup location: forge provider login. Verification: Configured from official documentation; not tested in this client.

Install ForgeCode, run forge provider login, select OpenAI-compatible, then enter endpoint/key and select a model.

Provider: OpenAI Compatible / Chat Completions
URL: https://modelfold.ai/v1
API Key: YOUR_MODELFOLD_API_KEY
Model ID: gpt-6-1-sol

Claude conversations through the OpenAI-compatible route do not use the native Claude prompt cache. For long tasks, consider a model with automatic caching or a client with native Anthropic support.

Official configuration reference.

MiMo Code

Setup location: TUI → Custom Provider. Verification: Configured from official documentation; not tested in this client.

Install @xiaomi-mimo/cli and add a Custom Provider in the TUI with URL, key and model.

Provider: OpenAI Compatible / Chat Completions
URL: https://modelfold.ai/v1
API Key: YOUR_MODELFOLD_API_KEY
Model ID: gpt-6-1-sol

Claude conversations through the OpenAI-compatible route do not use the native Claude prompt cache. For long tasks, consider a model with automatic caching or a client with native Anthropic support.

Official configuration reference.

ShellGPT

Setup location: Environment / ~/.config/shell_gpt/.sgptrc. Verification: Configured from official documentation; not tested in this client.

Install shell-gpt and set API_BASE_URL, OPENAI_API_KEY and DEFAULT_MODEL before running sgpt.

export API_BASE_URL=https://modelfold.ai/v1
export OPENAI_API_KEY="$MODELFOLD_API_KEY"
export DEFAULT_MODEL=gpt-6-1-sol
sgpt "Reply with exactly OK"

Claude conversations through the OpenAI-compatible route do not use the native Claude prompt cache. For long tasks, consider a model with automatic caching or a client with native Anthropic support.

Official configuration reference.

Trae Agent（开源 CLI）

Setup location: trae_config.yaml. Verification: Configured from official documentation; not tested in this client.

Install the repository CLI and configure base_url/api_key on its selected provider; its config is separate from TRAE IDE.

model_providers:
  modelfold:
    provider: anthropic
    base_url: https://modelfold.ai/anthropic
    api_key: YOUR_MODELFOLD_API_KEY
models:
  trae_agent_model:
    model_provider: modelfold
    model: claude-opus-5-5
    max_tokens: 8192

For Claude, prefer the native Anthropic route. Cache savings require eligible repeated prefixes and client cache support; a custom endpoint alone does not guarantee a cache hit.

Official configuration reference.

Coro Code

Setup location: Shell environment. Verification: Configured from official documentation; not tested in this client.

Install using the official guide, merge this configuration at the stated location, then select the ModelFold model.

export OPENAI_BASE_URL=https://modelfold.ai/v1
export OPENAI_API_KEY="$MODELFOLD_API_KEY"

Claude conversations through the OpenAI-compatible route do not use the native Claude prompt cache. For long tasks, consider a model with automatic caching or a client with native Anthropic support.

Official configuration reference.

Mini-Kode

Setup location: Shell environment. Verification: Configured from official documentation; not tested in this client.

Install using the official guide, merge this configuration at the stated location, then select the ModelFold model.

export MINIKODE_BASE_URL=https://modelfold.ai/v1
export OPENAI_API_KEY="$MODELFOLD_API_KEY"

Claude conversations through the OpenAI-compatible route do not use the native Claude prompt cache. For long tasks, consider a model with automatic caching or a client with native Anthropic support.

Official configuration reference.

Every Code

Setup location: Shell environment. Verification: Configured from official documentation; not tested in this client.

Install using the official guide, merge this configuration at the stated location, then select the ModelFold model.

export OPENAI_BASE_URL=https://modelfold.ai/v1
export OPENAI_API_KEY="$MODELFOLD_API_KEY"

Claude conversations through the OpenAI-compatible route do not use the native Claude prompt cache. For long tasks, consider a model with automatic caching or a client with native Anthropic support.

Official configuration reference.

RA.Aid

Setup location: Shell environment. Verification: Configured from official documentation; not tested in this client.

Install using the official guide, merge this configuration at the stated location, then select the ModelFold model.

export OPENAI_API_BASE=https://modelfold.ai/v1
export OPENAI_API_KEY="$MODELFOLD_API_KEY"

Claude conversations through the OpenAI-compatible route do not use the native Claude prompt cache. For long tasks, consider a model with automatic caching or a client with native Anthropic support.

Official configuration reference.

GitClaw

Setup location: Shell environment. Verification: Configured from official documentation; not tested in this client.

Install using the official guide, merge this configuration at the stated location, then select the ModelFold model.

export OPENAI_BASE_URL=https://modelfold.ai/v1
export OPENAI_API_KEY="$MODELFOLD_API_KEY"

Claude conversations through the OpenAI-compatible route do not use the native Claude prompt cache. For long tasks, consider a model with automatic caching or a client with native Anthropic support.

Official configuration reference.

NanoClaw

Setup location: .env. Verification: Configured from official documentation; not tested in this client.

Install using the official guide, merge this configuration at the stated location, then select the ModelFold model.

ANTHROPIC_BASE_URL=https://modelfold.ai/anthropic
ANTHROPIC_AUTH_TOKEN=YOUR_MODELFOLD_API_KEY

For Claude, prefer the native Anthropic route. Cache savings require eligible repeated prefixes and client cache support; a custom endpoint alone does not guarantee a cache hit.

Official configuration reference.

Crab Code

Setup location: ~/.crab/config.toml. Verification: Configured from official documentation; not tested in this client.

Install using the official guide, merge this configuration at the stated location, then select the ModelFold model.

api_provider = "anthropic"
base_url = "https://modelfold.ai/anthropic"
model = "claude-opus-5-5"
api_key = "YOUR_MODELFOLD_API_KEY"

For Claude, prefer the native Anthropic route. Cache savings require eligible repeated prefixes and client cache support; a custom endpoint alone does not guarantee a cache hit.

Official configuration reference.

Ante

Setup location: ~/.ante/catalog.json. Verification: Configured from official documentation; not tested in this client.

Install using the official guide, merge this configuration at the stated location, then select the ModelFold model.

{
  "providers": {
    "modelfold": {
      "base_url": "https://modelfold.ai/v1",
      "wire_style": "OpenAiCompatible",
      "credential": "env:MODELFOLD_API_KEY"
    }
  }
}

Add the exact model ID in the catalog model section; this is a provider fragment.

Claude conversations through the OpenAI-compatible route do not use the native Claude prompt cache. For long tasks, consider a model with automatic caching or a client with native Anthropic support.

Official configuration reference.

PicoClaw

Setup location: ~/.picoclaw/config.json. Verification: Configured from official documentation; not tested in this client.

Install using the official guide, merge this configuration at the stated location, then select the ModelFold model.

{
  "model_list": [
    {
      "model_name": "modelfold",
      "model": "openai/gpt-6-1-sol",
      "api_base": "https://modelfold.ai/v1",
      "api_key": "YOUR_MODELFOLD_API_KEY"
    }
  ]
}

Claude conversations through the OpenAI-compatible route do not use the native Claude prompt cache. For long tasks, consider a model with automatic caching or a client with native Anthropic support.

Official configuration reference.

Zap

Setup location: config.toml. Verification: Configured from official documentation; not tested in this client.

Install using the official guide, merge this configuration at the stated location, then select the ModelFold model.

[providers.modelfold]
kind = "openai"
model = "gpt-6-1-sol"
base_url = "https://modelfold.ai/v1/chat/completions"

Set the provider credential using the official authentication workflow; base_url is the complete request URL.

Claude conversations through the OpenAI-compatible route do not use the native Claude prompt cache. For long tasks, consider a model with automatic caching or a client with native Anthropic support.

Official configuration reference.

Nanocoder

Setup location: agents.config.json. Verification: Configured from official documentation; not tested in this client.

Install following the official guide, merge this configuration into the stated file, then select ModelFold.

{
  "nanocoder": {
    "providers": [
      {
        "name": "ModelFold",
        "sdkProvider": "anthropic",
        "baseUrl": "https://modelfold.ai/anthropic/v1",
        "apiKey": "${MODELFOLD_API_KEY}",
        "models": [
          "claude-opus-5-5"
        ],
        "maxOutputTokens": 8192
      }
    ]
  }
}

For Claude, prefer the native Anthropic route. Cache savings require eligible repeated prefixes and client cache support; a custom endpoint alone does not guarantee a cache hit.

Official configuration reference.

Plandex（自托管）

Setup location: plandex models custom. Verification: Configured from official documentation; not tested in this client.

Install following the official guide, merge this configuration into the stated file, then select ModelFold.

{
  "providers": [
    {
      "name": "modelfold",
      "baseUrl": "https://modelfold.ai/v1",
      "apiKeyEnvVar": "MODELFOLD_API_KEY"
    }
  ],
  "models": [
    {
      "modelId": "modelfold-gpt",
      "publisher": "ModelFold",
      "description": "ModelFold coding",
      "defaultMaxConvoTokens": 64000,
      "maxTokens": 128000,
      "maxOutputTokens": 8192,
      "reservedOutputTokens": 8192,
      "preferredOutputFormat": "xml",
      "providers": [
        {
          "provider": "custom",
          "customProvider": "modelfold",
          "modelName": "gpt-6-1-sol"
        }
      ]
    }
  ]
}

Custom providers require self-hosting. Cloud BYO Key only permits built-in providers.

Claude conversations through the OpenAI-compatible route do not use the native Claude prompt cache. For long tasks, consider a model with automatic caching or a client with native Anthropic support.

Official configuration reference.

Claw Code Agent (HarnessLab)

Setup location: 启动环境. Verification: Configured from official documentation; not tested in this client.

Install from the official project, configure the following settings and select the specified model.

export OPENAI_BASE_URL="https://modelfold.ai/v1"
export OPENAI_API_KEY="$MODELFOLD_API_KEY"
export OPENAI_MODEL="gpt-6-luna"

This is distinct from ultraworkers Claw Code.

Claude conversations through the OpenAI-compatible route do not use the native Claude prompt cache. For long tasks, consider a model with automatic caching or a client with native Anthropic support.

Official configuration reference.

Free Code

Setup location: 启动环境. Verification: Configured from official documentation; not tested in this client.

Install from the official project, configure the following settings and select the specified model.

export ANTHROPIC_BASE_URL="https://modelfold.ai/anthropic"
export ANTHROPIC_API_KEY="$MODELFOLD_API_KEY"
export ANTHROPIC_MODEL="claude-haiku-5-5"

Third-party project, not the official Anthropic Claude Code.

For Claude, prefer the native Anthropic route. Cache savings require eligible repeated prefixes and client cache support; a custom endpoint alone does not guarantee a cache hit.

Official configuration reference.

Pool (Poolside)

Setup location: 启动环境. Verification: Configured from official documentation; not tested in this client.

Install from the official project, configure the following settings and select the specified model.

export POOLSIDE_STANDALONE_BASE_URL="https://modelfold.ai"
export POOLSIDE_API_KEY="$MODELFOLD_API_KEY"
export POOLSIDE_STANDALONE_MODEL="gpt-6-luna"
pool

The client appends /v1; use standalone mode.

Claude conversations through the OpenAI-compatible route do not use the native Claude prompt cache. For long tasks, consider a model with automatic caching or a client with native Anthropic support.

Official configuration reference.

zot

Setup location: $ZOT_HOME/models.json. Verification: Configured from official documentation; not tested in this client.

Install from the official project, configure the following settings and select the specified model.

{
  "providers": {
    "modelfold": {
      "baseUrl": "https://modelfold.ai/anthropic",
      "api": "anthropic",
      "models": [
        {
          "id": "claude-haiku-5-5",
          "name": "ModelFold"
        }
      ]
    }
  }
}

Set MODELFOLD_API_KEY, then run zot --provider modelfold --model claude-haiku-5-5. Do not store credentials in models.json.

For Claude, prefer the native Anthropic route. Cache savings require eligible repeated prefixes and client cache support; a custom endpoint alone does not guarantee a cache hit.

Official configuration reference.

nanobot

Setup location: ~/.nanobot/config.json. Verification: Configured from official documentation; not tested in this client.

Install from the official project, configure the following settings and select the specified model.

{
  "providers": {
    "custom": {
      "apiKey": "${MODELFOLD_API_KEY}",
      "apiBase": "https://modelfold.ai/v1"
    }
  },
  "modelPresets": {
    "primary": {
      "provider": "custom",
      "model": "gpt-6-luna",
      "maxTokens": 4096,
      "contextWindowTokens": 65536
    }
  },
  "agents": {
    "defaults": {
      "modelPreset": "primary"
    }
  }
}

Run nanobot agent -m "Reply OK". For native Claude select Anthropic Messages in provider advanced options, with the /anthropic URL and a Claude model.

For Claude, prefer the native Anthropic route. Cache savings require eligible repeated prefixes and client cache support; a custom endpoint alone does not guarantee a cache hit.

Official configuration reference.

Deep Agents Code (LangChain)

Setup location: ~/.deepagents/config.toml. Verification: Configured from official documentation; not tested in this client.

Install from the official project, configure the following settings and select the specified model.

[models.providers.anthropic]
base_url = "https://modelfold.ai/anthropic"
api_key_env = "MODELFOLD_API_KEY"
models = ["claude-haiku-5-5"]

Install official deepagents-code; set the environment variable and run dcode --model anthropic:claude-haiku-5-5.

For Claude, prefer the native Anthropic route. Cache savings require eligible repeated prefixes and client cache support; a custom endpoint alone does not guarantee a cache hit.

Official configuration reference.