Connect DeepSeek Harness to Scenepond
Add Scenepond as a custom model API for coding and repository tasks.
Official integration documentation- API base URL
https://scenepond-api-production.up.railway.app/v1- API protocol
- OpenAI Chat Completions
- Model ID
deepseek-flash· Use the exact catalog ID
DeepSeek Harness provides a web interface for coding tasks in your workspace. Add Scenepond as a custom provider, then select a model for the session.
This guide assumes the client is installed and can start locally. You also need a Scenepond account, an enabled chat model, an API key with access, and sufficient balance for your first request.
Client installation and first launch ↗Set up with your coding agent
Copy this setup brief to the coding agent you already use. It contains the endpoint and instructions, without your API key. You can also follow the manual steps below.
Help me connect DeepSeek Harness to Scenepond using its custom provider. Scenepond guide: https://open.scenepond.ai/en/integrations/deepseek-harness?model=deepseek-flash Official integration documentation: https://deepseek-harness.github.io/deepseek-harness/en/guide/providers API base URL: https://scenepond-api-production.up.railway.app/v1 Model ID: deepseek-flash API protocol: OpenAI Chat Completions Inspect the installed client version and existing configuration first. Preserve other providers, agents, channels, and settings. Confirm that my API key permits the selected model. Verify tool support and actual context limits; do not invent capabilities. I will set the raw key in SCENEPOND_API_KEY or the client’s secure credential field. Do not ask me to paste it into this conversation, print it, commit it, or embed it in browser code. Follow the Scenepond guide for this client’s provider syntax and secret handling. Use the /v1 base URL and chat completions protocol with Scenepond credentials, rather than upstream provider credentials. Validate configuration locally first. Ask before making a paid API request. If a model or credential is missing, explain what I need to configure. Do not retry requests automatically. Report the changed settings, how to start a fresh session, and what has and has not been verified.
1. Prepare your model and API key
Copy an enabled model ID from the catalog and create a Scenepond API key with access to it. Agent tasks need a model with verified tool-calling support; check context limits and supported protocols before choosing.
Use deepseek-flash from the catalog and confirm that your API key has access. These instructions configure a chat endpoint; image and video tools require their own setup.
2. Configure the custom provider
In DeepSeek Harness, open Settings → Models → Add model provider, then switch to Custom model API. Use the custom provider for Scenepond so the endpoint and model list come from your gateway.
- Provider ID
- scenepond
- Display name
- Scenepond
- Base URL
https://scenepond-api-production.up.railway.app/v1- API protocol
- OpenAI Chat Completions
- API key
- Your Scenepond API key, without the Bearer prefix
- Model ID
deepseek-flash
Add the exact model ID manually, or use Fetch available models and select a model permitted by your API key. Save the provider; use verified model limits and input types when setting optional capabilities.
3. Choose a workspace and model
Select your workspace, choose the configured Scenepond model, and start a new session. A session that has already sent a request keeps its recorded model. Model changes apply on the next request without restarting the server.
DeepSeek Harness is in developer preview. This guide configures its custom chat provider; it does not imply that a DeepSeek model is already available in the Scenepond catalog.
4. Confirm the response and usage
Confirm that the client has selected your Scenepond provider and exact model ID. Send this short message in a new session; the request uses your account balance.
Reply READY without using tools.Check that a text reply arrives, then open Scenepond usage records and match the model and request time. Review the actual settled charge. A text reply verifies basic chat access; test tools and reasoning separately only when the model supports them.
Common setup issues
- Missing key or rejected credentials
- Check that the running client can read the raw key and that the Scenepond key is enabled. Keep the secret out of diagnostic output.
- Model unavailable
- Check the exact catalog ID, key permissions, and balance. For OpenClaw, the model reference also needs the scenepond/ prefix.
- Session uses the old model
- Save the custom provider, select its model in the workspace, and start a new session. A session that has sent a request keeps its recorded model.
- Chat works, tools fail
- Check the selected model’s tool support and the client’s tool settings. Successful text chat does not establish support for every agent capability.