DeepSeek Harness
DeepSeek Harness Web UI Tutorial: Models, Workspaces, and Agent Tasks
Use the verified Web UI workflow to configure a model, select a workspace, run a task, and understand approval prompts.
What you will learn
- Model configuration and workspace selection are separate setup steps.
- Begin with a read-only task to validate the complete path.
- Record profile and permission context when comparing runs.
Before you start
- Basic HTTP and API knowledge
Leave with a concrete implementation checklist and a testable starting point.
Key takeaways
- Model configuration and workspace selection are separate setup steps.
- Begin with a read-only task to validate the complete path.
- Record profile and permission context when comparing runs.
Configure before composing
The Web UI guide puts model configuration first: open Settings → Models, enter a DeepSeek API key, and save it. The route becomes usable without restarting the server, so you can correct a credential or provider setting in the UI.
Next choose a workspace. The process directory is only the default filesystem location; a fresh Web UI has no selected workspace, and the session composer remains unavailable until you explicitly add one.
Run a bounded task
Start with a read-oriented request such as “Summarize this repository and identify its main packages.” This checks that the model, workspace, session log, and tool registry are connected before you permit edits.
The guide says the agent can read and edit files, run commands, delegate work, and maintain a plan. Approval prompts are part of the active permission policy, so the same task can require different human decisions under another profile.
What to record
For a reproducible tutorial, record the dsh version, profile, workspace path, model route, and whether an approval was accepted or denied. These details explain why two apparently identical prompts can produce different tool traces.
Keep the Web UI as an operator surface, not as a substitute for source-level tests. When a workflow matters, reproduce it in the CLI or Python SDK and keep the resulting session artifacts.
Decision guide
| Criterion | Option A | Option B |
|---|---|---|
| Best when | You need predictable behavior and easy auditing | You need adaptive optimization and have reliable telemetry |
| Main risk | May leave performance on the table | Can become difficult to explain or debug |
Implementation steps
- 1
Start `dsh web` and open the printed URL.
- 2
Save the DeepSeek model credential in Settings → Models.
- 3
Choose the project directory as the workspace.
- 4
Run a read-only task and inspect approvals before enabling edits.
Copy-ready example
Settings -> Models -> save credential
Choose workspace -> Start session -> task -> approval -> tool resultFrequently asked questions
Why is the composer disabled?
The Web UI guide requires a workspace to be selected before the session composer becomes available.
Sources
- DeepSeek Harness Web UI guideSource checked 2026-08-27