Choose an API setup guide for your coding harness

Clients and setup paths

• DeepSeek Harness

  Connect DeepSeek Harness to ApiClo through Settings → Models → Add model provider → Custom model API. Choose OpenAI Chat Completions and https://apiclo.com/v1. Harness is the client; your selected ApiClo model and tariff determine model usage. This setup does not connect the DeepSeek chat website's search service.

• Oh My Pi (OMP)

  Oh My Pi uses ~/.omp/agent/models.yml for a custom ApiClo provider. Set baseUrl to https://apiclo.com/v1, api to openai-completions and apiKey to the environment variable name APICLO_API_KEY. OMP and Pi have different configuration files and key syntax.

• Pi

  Connect the Pi coding agent to ApiClo in ~/.pi/agent/models.json. Use baseUrl https://apiclo.com/v1, api openai-completions and apiKey $APICLO_API_KEY. Open /model to reload the file and choose the custom provider. This guide concerns the Pi coding agent, not another product named Pi.

• Kimi Code

  Kimi Code can configure an OpenAI-compatible provider with an ApiClo /v1 endpoint. ApiClo's Kimi Code integration is still undergoing compatibility testing: treat this as a setup candidate, not a validated autonomous workflow. Kimi Code credentials and the Kimi chat/search service are separate.

• Orca

  Use ApiClo with Orca by configuring the coding agent that Orca launches: Codex CLI, Claude Code, Pi. Orca orchestrates agent sessions and worktrees; the child agent owns its API endpoint, credentials and model configuration.

• Crush

  Add ApiClo to current Crush using an openai-compat provider in the global crushrc file. Use https://apiclo.com/v1 and a key from APICLO_API_KEY. Current Crush configuration uses its built-in Bash commands; older JSON examples belong to a different configuration format.

• Goose

  Goose supports custom OpenAI-compatible providers for ApiClo. Configure the provider in Settings → Models or through goose configure, save the ApiClo key using its credential flow, and add the exact model IDs. Verify whether your provider screen expects a full API URL or a host plus path.

• Codex CLI / IDE

  Responses API; user-level config.toml and env_key.

• Claude Code

  Anthropic Messages; root URL and ANTHROPIC_AUTH_TOKEN.

• VS Code: Cline / Roo Code / Kilo Code

  Extension provider settings; /v1 and exact model IDs.

• Cursor

  Custom API setup; check Cursor-specific model IDs and feature restrictions.

• Trae

  Connect Trae to ApiClo through Settings → Models → Add Model → Custom Config when that option is available in your version and account. Choose the API format before entering the URL: OpenAI-compatible Chat Completions and Anthropic Messages use different request paths.

• Continue

  config.yaml: provider, apiBase, apiKey and model roles.

• VS Code native BYOK

  Custom Endpoint model configuration; tool and vision capabilities need checking.

• OpenClaw

  Separate OpenAI and Anthropic provider modes; use the matching URL and model input types.

• Qwen Code

  Use an OpenAI-compatible client mode with OPENAI_* variables where supported.

• Aider

  OpenAI-compatible endpoint and openai/ model selector prefix.

• Open WebUI / LibreChat

  Self-hosted chat UI; configure the server's API connection and credentials.

• Hermes Agent

  Managed workspace; hosting and model-token usage are separate charges.

• Gemini CLI

  Conditional: direct ApiClo setup requires a build that accepts a compatible custom endpoint.

• ApiClo Desktop / Local Router

  Local routing option; check the currently available build and client requirements.

Before you start

• Choose the exact client and version; similar product names do not imply identical configuration.
• Create an ApiClo key and check model capabilities and current input/output/cache rates.

Steps

1. Choose the protocol

   OpenAI-compatible Base URL is usually https://apiclo.com/v1. Claude Code uses https://apiclo.com because it adds /v1/messages. A field asking for a full request URL needs the actual endpoint path.

2. Verify a real request

   Test text, then tools and images independently. Inspect Logs for the exact model and token use before parallel or unattended tasks.

3. Compare the full token cost

   Read input, output, cache-read and cache-write rates. The 40/40/20 example assumes 1M total tokens: 0.4 × input + 0.4 × read + 0.2 × output; initial writes are excluded. Use actual Logs for your own average.

Check the result

Follow the selected guide and compare the successful request with ApiClo Logs. Documentation availability is not a claim that every version, model or feature has passed integration tests.

Troubleshooting

• DeepSeek Harness is a client; DeepSeek chat's web search is a separate service.
• Kimi Code is a coding client; using it with ApiClo does not include a Kimi subscription or Kimi search tools.
• GLM, Kimi and DeepSeek model availability must be checked in the current ApiClo catalog; a client's name does not establish that those models are sold here.

Compatibility and billing

ApiClo is independent of the listed clients and AI providers. Kimi Code remains under testing; Orca uses a child agent; Gemini CLI needs compatible custom-endpoint support. API balance, client subscriptions and managed hosting are separate. No cache-hit rate or universally lowest price is guaranteed.