DeepSeek Harness, with Pareto in the model slot

Q: Which API protocol should I choose?

OpenAI Chat Completions, with https://api.unbiased.ai/v1 as the Base URL. That is the setup we ran on October 20, 2026: DeepSeek Harness 0.2.1-rc.2 read a failing Python test, fixed the file with its edit tool and reran the test to green. Unbiased also accepts OpenAI Responses and Anthropic Messages requests, but one dsh provider speaks one protocol, so another protocol means a second provider, and we have not tested those. Format compatibility is not a promise that every feature works, so tell us if one doesn’t.

DeepSeek Harness builds the agent out of plugins, and the model is one more setting. Add Unbiased as a Custom model API provider in Settings → Models, type pareto as the model, and the tools, workspace and sessions carry on as before. For each request, Pareto selects frontier and open models and blends their outputs into one answer.

Choose Personal · $10/month See the setup →

Tested live with DeepSeek Harness 0.2.1-rc.250M tokens a week

~/proj · dsh --profile headless

export UNBIASED_API_KEY=your-unbiased-api-key

dsh --profile headless --json "test_chunk.py fails. Fix the bug in chunk.py, then run python3 test_chunk.py to confirm."

session started · cwd ~/proj

I’ll start by looking at the files.

bash(ls -la)

read(chunk.py)

read(test_chunk.py)

The bug is in the loop bound: len(items) - size + 1 stops before the final partial chunk, so [5] is dropped.

edit(chunk.py)

- for i in range(0, len(items) - size + 1, size):+ for i in range(0, len(items), size):

bash(python3 test_chunk.py)

Fixed and confirmed. The old range stopped one step early, so chunk([1,2,3,4,5], 2) lost its last chunk. It now runs to len(items), and the test prints ok.

unbiased / pareto · openai-completionsturn_end: completed

A demonstration of DeepSeek Harness on Pareto, recreated from a real run. Not a screenshot.

Add Pareto as a Custom model API

DeepSeek Harness takes any OpenAI- or Anthropic-compatible endpoint as a Custom model API provider. Fill in the form in the Web UI, or write the same settings into your profile’s cordis.patch.yml; the Web UI writes to that file too.

Get an Unbiased API key
Choose a plan, then create a key in the Unbiased platform. New accounts are reviewed and approved by a person before the key works.
Add a Custom model API provider
Start the Web UI with npx @deepseek-ai/dsh web, open Settings → Models, choose Add model provider and switch the card to Custom model API. Use unbiased as the Provider ID (it can’t be renamed later), https://api.unbiased.ai/v1 as the Base URL, OpenAI Chat Completions as the API protocol, and paste your key.
Type pareto by hand
Unbiased does not publish a model list, so Fetch available models won’t find it. Choose Add model, type pareto and choose Create provider. DeepSeek’s docs say hand-added ids work just the same.
Pick it and give it a task
Select pareto in the model picker; it becomes the default for new sessions. Model changes apply on the next request, so the server needs no restart. For one-shot jobs from a terminal, run dsh --profile headless "…" with the file settings below.

Tested on October 20, 2026DeepSeek Harness 0.2.1-rc.2 on Pareto (pareto). It read the failing Python test, fixed the off-by-one in chunk.py with its edit tool and reran the test to green. Run record.

Web UI: Custom model API

# Settings → Models → Add model provider
Provider ID:   unbiased
Base URL:      https://api.unbiased.ai/v1
API protocol:  OpenAI Chat Completions
API key:       your-unbiased-api-key
Models:        pareto
# Add the model by hand; skip Fetch available models

headless/cordis.patch.yml

# $DSH_HOME/profiles/headless/cordis.patch.yml
- id: llm-pi-ai
  config:
    providers:
      unbiased:
        apiKeyEnv: UNBIASED_API_KEY
        api: openai-completions
        baseURL: https://api.unbiased.ai/v1
        models:
          - id: pareto
- id: agent-default-model
  config:
    provider: unbiased
    model: pareto

Shell

export UNBIASED_API_KEY=your-unbiased-api-key
dsh --profile headless "test_chunk.py fails. Fix the bug in chunk.py, then run python3 test_chunk.py to confirm."
# Add --json to stream each tool call as an event

Each profile has its own file: the Web UI started with dsh web uses profiles/web. An entry in cordis.patch.yml replaces that plugin’s whole config, so keep any other providers you have inside the same llm-pi-ai entry.

DeepSeek Harness stays. The model changes.

DeepSeek Harness describes itself as everything-is-a-plugin. Changing the model edits one plugin’s settings and leaves the others where they were.

Tools, plugins and the workspace you choseUnchanged
Web UI, headless, ACP and SDK profilesSame dsh launcher
Sessions you already startedKeep the model they began with
The Effort menuHidden until you add reasoning levels
Provider and modelunbiased / pareto
What you payFrom $10 a month

One model string, several models behind it

Pareto is Unbiased’s blended AI model. On each request it selects frontier and open models and blends their outputs into one answer, so DeepSeek Harness talks to one model string, pareto, and never has to pick.

“When the frontier improves, Pareto improves; when frontier prices fall, you should benefit.” From the pricing page.

How a Pareto answer gets made →

An illustration, not a wiring diagram. The blend includes frontier and open models, and its composition can change.

What an agent task costs on Pareto

Agents make many model calls per task, so the price of a finished task matters more than the price of a token. The model card publishes both for Pareto on two public agent benchmarks.

Terminal-Bench 4.0Tasks an agent completes in a terminal: $0.48/task
DeepSWE v1.1Changes an agent makes in a real codebase: $0.24/task

Pareto 26.10, preliminary October 1, 2026 runs that may change; denominators and the cost method still require confirmation. Cost is the reported mean USD per task at API rates, not an estimate for your workload. Model card and method.

Pay for DeepSeek Harness by the month

One API key works across DeepSeek Harness and any other tool you point at Pareto. Pick a monthly allowance, or pay per token on Team.

Personal

$10/month

50M tokens a week
20 requests per minute
For one person

Choose Personal

Personal Max

$100/month

600M tokens a week
60 requests per minute
For one person

Choose Personal Max

Team

Pay as you go

Prepaid credits at published rates
300 requests per minute
No weekly token cap

Get started with Team

Subscriptions are for one person and are not unlimited. New API accounts are reviewed by a person before access turns on, and commercial or production use needs a separate written agreement. Rates and limits are on Pricing.

DeepSeek Harness on Pareto, questions answered

Which API protocol should I choose?

OpenAI Chat Completions, with https://api.unbiased.ai/v1 as the Base URL. That is the setup we ran on October 20, 2026: DeepSeek Harness 0.2.1-rc.2 read a failing Python test, fixed the file with its edit tool and reran the test to green. Unbiased also accepts OpenAI Responses and Anthropic Messages requests, but one dsh provider speaks one protocol, so another protocol means a second provider, and we have not tested those. Format compatibility is not a promise that every feature works, so tell us if one doesn’t.

Why can’t Fetch available models find pareto?

Unbiased does not serve a model list, so discovery fails or lists nothing. Instead, choose Add model, type pareto and choose Create provider. The DeepSeek Harness docs treat discovery as a convenience and say ids added by hand work just the same.

Do I need a DeepSeek account or API key?

No. Pareto requests go to Unbiased with your Unbiased key, and your Unbiased plan pays for them. The DeepSeek card on the Models page can stay empty. DeepSeek Harness itself is open source under the MIT license.

Can I switch back to a DeepSeek model?

Yes. Add a DeepSeek key and pick one of its models in the model picker; that becomes the default for new sessions. A session that has already sent a request keeps the model recorded in its own log, so start a new session to switch.

Will this setup keep working as dsh changes?

DeepSeek Harness is a developer preview, and its README warns of compatibility-breaking changes. The settings here worked with 0.2.1-rc.2. If a field moves in a later release, follow DeepSeek’s Configure models guide.

Is Pareto made by the DeepSeek Harness team?

No. Pareto is Unbiased’s model. DeepSeek Harness is made by DeepSeek, and Unbiased is not affiliated with or endorsed by them. You keep DeepSeek Harness and point it at the Unbiased API.

What happens when I hit the weekly allowance?

Requests over your plan’s limits return an HTTP 429 error. Personal includes 50M tokens a week and 20 requests per minute. Personal Max raises that to 600M a week and 60 requests per minute for $100/month. Team uses prepaid pay-as-you-go credits with no weekly token cap.

Do I choose between models?

No. You send every request to one model string, pareto. Pareto selects frontier and open models for each request and blends their outputs into one answer.

Is my code kept?

Personal and Personal Max are not our zero data retention tier: prompts and responses may be kept up to 30 days, as section 5(c) of the Terms explains. For work that must not be kept, ask about the Zero Data Retention API on the Private AI page.

Pareto works in other tools too

The same key and the same model string go into each of these. See them all.

Product names belong to their owners. Except where a guide says otherwise, Unbiased is not affiliated with or endorsed by the makers of the tools on this page.

Hand DeepSeek Harness the test that keeps failing

One provider, one hand-typed model id and the plugins you already run. Pareto starts at $10 a month for 50M tokens a week.