> ## Documentation Index > Fetch the complete documentation index at: https://docs.eigi.ai/llms.txt > Use this file to discover all available pages before exploring further. # Connect Your Own Harness > Use OpenClaw, Harness, or any OpenAI-compatible Responses API with eigi.ai ## Overview You can connect your existing AI harness to eigi.ai instead of rebuilding it from scratch. This works with: * OpenClaw * Harness * Any custom harness that exposes an OpenAI-compatible Responses API In simple terms, you keep your own model stack running where you want, then point eigi.ai to that secure HTTPS endpoint. This is the right setup if you already have your own orchestration, tools, prompts, guardrails, or model routing and want to use it inside eigi.ai. *** ## Before You Start Make sure you have these three things ready: * A working harness with an OpenAI-compatible Responses API * A public HTTPS URL for that harness * A secret API key or token that protects access to the harness If your harness is only running on your laptop or on a private server today, that is fine. You can still connect it by exposing it safely over HTTPS. *** ## How The Flow Works Pick the system you want eigi.ai to call. This can be OpenClaw, Harness, or your own custom harness as long as it exposes an OpenAI-compatible Responses API. Run that harness on your own laptop, another local machine, a private server, AWS, or any cloud you trust. The choice depends on your security and compliance requirements. Make the harness reachable on the internet through a secure HTTPS URL. Most teams do this with a public server, `ngrok`, or Cloudflare Tunnel. Create an API key, bearer token, or similar secret so eigi.ai can authenticate when calling your harness. In eigi.ai Agent360, while creating or editing an agent, choose **Custom LLM** and enter the base URL, API key, and model name for your harness. Save the connection, run a test request, and confirm eigi.ai can reach your harness successfully. *** ## Step 1: Choose A Harness Your harness is the service that receives the model request and returns the AI response. You can use: * **OpenClaw** if you already use OpenClaw workflows * **Harness** if your team already has that runtime in place * **A custom harness** if your application already exposes an OpenAI-compatible Responses API If you are unsure whether your setup is compatible, check whether your service already accepts OpenAI-style requests and returns OpenAI-style responses. *** ## Step 2: Decide Where To Host It You can host your harness in any location that matches your security needs. | Hosting option | Best for | Notes | | ----------------------------------------- | ------------------------------ | ------------------------------------------------------------------ | | **Your laptop or desktop** | Quick testing | Fastest way to validate the setup, but your machine must stay on | | **Another machine on your local network** | Internal testing | Useful when a dedicated machine is available in your office or lab | | **AWS, GCP, Azure, or any cloud VM** | Production use | Best for stable uptime and easier team access | | **Private company infrastructure** | Security-sensitive deployments | Good when data must remain inside your own environment | If your team has strict security rules, start with your approved internal or cloud environment instead of a laptop-based setup. *** ## Step 3: Make Your Harness Reachable Over HTTPS eigi.ai needs a secure HTTPS URL to call your harness. If your harness is already deployed on a public server with HTTPS, you can use that URL directly. If it is only running locally or on a private network, you need a secure tunnel. ### What is ngrok? `ngrok` is a tunneling tool that gives your local service a temporary public URL. It is useful when: * You want to test quickly from your local machine * You do not want to deploy to the cloud yet * You need a short-lived public HTTPS endpoint for development ### When should you use Cloudflare Tunnel? Cloudflare Tunnel is a better fit when: * You want a more stable public hostname * You want to use your own domain * You need a production-friendly tunnel option * Your team already uses Cloudflare for networking or security ### Choose the right option | Option | Use it when | Avoid it when | | --------------------------------- | -------------------------------------------- | ---------------------------------------- | | **ngrok** | You want the fastest local proof of concept | You need a long-term production endpoint | | **Cloudflare Tunnel** | You want a more durable and controlled setup | You only need a quick one-time test | | **Direct cloud HTTPS deployment** | You already run services in the cloud | You are still validating locally | *** ## Step 4: Install ngrok Or Cloudflare Tunnel If you are handing this to another agent, give it your **OS**, **local port**, and tunnel choice, then ask it to either run `ngrok http ` after ngrok install and auth setup, or run `cloudflared tunnel --url http://localhost:` after Cloudflare install, and return the final public HTTPS URL. ### Option A: Install ngrok Install `ngrok` from the official site: 1. Go to [ngrok.com/download](https://ngrok.com/download) 2. Download the version for your operating system 3. Install it using the instructions shown by ngrok 4. Create an ngrok account if required 5. Add your ngrok auth token following ngrok's setup instructions Once your harness is running locally on a port such as `3000` or `8000`, start the HTTPS tunnel: ```bash theme={null} ngrok http 8000 ``` ngrok will give you a public HTTPS URL such as: ```text theme={null} https://example-name.ngrok-free.app ``` Use that HTTPS URL as your harness base URL in eigi.ai. ### Option B: Install Cloudflare Tunnel Install Cloudflare Tunnel from the official Cloudflare documentation: 1. Go to [developers.cloudflare.com/cloudflare-one/connections/connect-networks/downloads](https://developers.cloudflare.com/cloudflare-one/connections/connect-networks/downloads/) 2. Install `cloudflared` for your operating system 3. Sign in to Cloudflare if your setup requires account authorization 4. Create a tunnel and map it to your local harness port For a quick local tunnel, Cloudflare commonly uses a command like this: ```bash theme={null} cloudflared tunnel --url http://localhost:8000 ``` Cloudflare will provide a public HTTPS hostname you can use as your harness base URL. Use only one public URL for the final configuration. If your ngrok or Cloudflare URL changes, update the saved base URL in eigi.ai. *** ## Step 5: Create A Secret For Authentication Do not expose your harness without protection. Create a secret value that eigi.ai will send with each request. Depending on your harness, this can be: * An API key in a header * A bearer token * A gateway secret managed by your own reverse proxy Recommended approach: 1. Generate a strong random secret 2. Save it in your harness configuration or proxy 3. Configure your harness to reject requests that do not include that secret 4. Copy the same secret into eigi.ai when you configure the **Custom LLM** settings for your agent Never put this secret in frontend code, screenshots, shared docs, or public repositories. *** ## Step 6: Add The Harness In eigi.ai Once your harness has a public HTTPS URL and an API secret, open eigi.ai and add it as a **Custom LLM** while creating or editing your agent in Agent360. In Agent360: 1. Open **Create Agent** or edit an existing agent 2. Go to the LLM configuration section 3. Choose **Custom LLM** as the provider 4. Enter your harness details 5. Save the agent Enter these values: * **Base URL**: the public HTTPS address for your harness * **API Key**: the secret token eigi.ai should send for authentication * **Model Name**: the model identifier your harness expects Examples: | Field | Example | | -------------- | ----------------------------------------------------- | | **Base URL** | `https://your-harness.example.com` | | **API Key** | `sk_live_example_secret` | | **Model Name** | `gpt-4.1`, `gpt-4o`, or your custom routed model name | You can do this from: * The eigi.ai dashboard * Our MCP-based workflow * Our CLI-based workflow The important part is the same in every case: save the **base URL**, **API key**, and **model name** for your external harness in the agent's **Custom LLM** configuration. *** ## Step 7: Test The Connection After saving the **Custom LLM** configuration: 1. Send a simple test request 2. Confirm the harness responds successfully 3. Check that the returned response format is valid 4. Verify that authentication is working correctly If the test works, your harness is now connected to eigi.ai. *** ## Quick Checklist * I chose a harness that supports an OpenAI-compatible Responses API * I hosted it on a machine I trust * I exposed it through a public HTTPS URL * I protected it with an API key or token * I added the base URL, API key, and model name in eigi.ai * I ran a successful test *** ## Common Problems ### eigi.ai cannot reach the harness Check that: * The URL is public and starts with `https://` * Your tunnel or server is still running * Your firewall allows inbound traffic * The saved base URL is correct ### Authentication fails Check that: * The secret in eigi.ai matches the secret expected by your harness * Your harness is reading the correct header * The secret has not expired or been rotated ### The harness responds, but the model call fails Check that: * The model name in eigi.ai matches the model name your harness expects * Your harness supports the required OpenAI-compatible Responses API behavior * Your harness is healthy and can reach its own underlying model provider *** ## Recommended Path For Most Teams If you are not sure where to begin, use this order: 1. Start with your harness on a local machine 2. Expose it temporarily with `ngrok` 3. Validate the connection from eigi.ai 4. Move it to Cloudflare Tunnel or a cloud HTTPS deployment for long-term use This keeps the first setup simple while still giving you a clean path to a more stable production deployment.