<!-- Generated from providers/deepseek.html. Do not edit by hand. -->

> Markdown twin of https://conduitllm.com/providers/deepseek.html
> Use the DeepSeek API in a free, open-source desktop app. Setup, peak and off-peak prices, thinking mode, tools, where your data goes, and current limits.

---
Provider guide

# Use the DeepSeek API in a desktop app

DeepSeek's API is one of the cheapest ways to use a frontier-class model, and its models think before they answer. Conduit is a free, open-source desktop app for your DeepSeek key: watch the reasoning, use MCP tools, switch to another provider mid-conversation, and keep every conversation on your own disk. DeepSeek details below were checked against DeepSeek's own documentation on 4 October 2026; if something has changed since, their documentation is right. Part of the [provider guides](https://conduitllm.com/providers.html).

## Set it up

1. **Get a key.** Sign in to the [DeepSeek platform](https://platform.deepseek.com/api_keys), create an API key, and top up your balance; DeepSeek bills from a prepaid balance.
2. **Pick DeepSeek.** In Conduit, open **Settings → Providers & keys** and choose **DeepSeek** from the **Provider** list.
3. **Save the key.** Paste it into **Provider secret** and choose **Save provider key**. It goes into your operating system's keychain, and the interface never sees it again.
4. **Pick a model.** Choose **Load models**, then pick `deepseek-flash` (fast and cheap) or `deepseek-v4-pro` from the **Model** list.

Keep a key for another provider too and Ctrl+Shift+P (⌘+Shift+P on macOS) switches the next answer to it in the same chat — useful for checking a cheap DeepSeek answer against another model.

## What works with DeepSeek

| Feature | With DeepSeek |
| --- | --- |
| Chat | Yes, streamed, with DeepSeek's reasoning shown above each answer. |
| MCP tools and built-in tools | Yes. A tool asks before it runs unless it is marked read-only. |
| Web search | Yes, through Conduit's own search (Exa by default, no key needed). DeepSeek has no search of its own. Off by default; asks before the first search. |
| Images in your messages | Not yet — see "What doesn't work yet". |
| Image generation | No. DeepSeek doesn't generate images; use OpenAI, Gemini, or OpenRouter for that. |
| Your documents | Not while DeepSeek is active. Index them with OpenAI, Gemini, OpenRouter, or Ollama. |
| Personal apps, Slides, workflows | Yes, like any other provider. |
| Token counts | Yes, per reply, including tokens served from DeepSeek's cache. |
| Cost in dollars | Not yet. Check your balance on the DeepSeek platform. |

## What it costs

Conduit is free and adds nothing. DeepSeek charges per million tokens, and every price halves off-peak. Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday, excluding Chinese public holidays.

| USD per million tokens | deepseek-flash | deepseek-v4-pro |
| --- | --- | --- |
| Input, cache hit | $0.003 off-peak, $0.006 peak | $0.022 off-peak, $0.044 peak |
| Input, cache miss | $0.15 off-peak, $0.30 peak | $0.66 off-peak, $1.32 peak |
| Output | $0.60 off-peak, $1.20 peak | $1.98 off-peak, $3.96 peak |

DeepSeek's caching is on for everyone: when a request repeats the start of an earlier one, as each new message in a chat does, the repeated part is billed at the cache-hit price. Both models take up to 1 million tokens of context. Whether paying per token beats a subscription depends on how much you use it; the [API vs subscription calculator](https://conduitllm.com/api-vs-subscription.html) works it out from your own numbers.

## Where your data goes

Your conversations are stored on your own disk, and Conduit sends each request straight from your machine to DeepSeek, with no Conduit server in between. DeepSeek's [privacy policy](https://cdn.deepseek.com/policies/en-US/deepseek-privacy-policy.html) says: "we directly collect, process and store your Personal Data in People's Republic of China." If your work has data-residency rules, read DeepSeek's [platform terms](https://cdn.deepseek.com/policies/en-US/deepseek-open-platform-terms-of-service.html) first, or use DeepSeek models through Together AI or Fireworks AI, which serve them from their own infrastructure under their own terms. Turn on local-only mode and Conduit refuses DeepSeek and every other cloud provider.

## What doesn't work yet

- **No images** — Conduit treats DeepSeek as text-only and leaves attached images out with a note, although DeepSeek's flash model now accepts images.
- **Context gauge assumes 128K** — Conduit's gauge, and the automatic compaction that follows it, treat DeepSeek models as having 128,000 tokens of context, so long chats are summarised much earlier than DeepSeek's 1 million requires.
- **No dollar costs** — token counts only; your spend is on the DeepSeek platform.
- **No thinking switch** — DeepSeek's thinking stays on; Conduit has no control to turn it off or change its effort.
- **No custom endpoint** — DeepSeek's address is fixed; for a proxy in front of it, use the OpenAI Compatible provider.

[Download Conduit](https://github.com/runningpixels/conduit/releases)

Other providers: [all provider guides](https://conduitllm.com/providers.html), including [OpenRouter](https://conduitllm.com/providers/openrouter.html), which also serves DeepSeek models. Why people bring their own key: [pay per token, not per seat](https://conduitllm.com/use-cases/bring-your-own-key.html).

## Questions

**Does Conduit charge anything on top of DeepSeek?**

No. Conduit is free and open source, with no account and no subscription. Requests go from your machine to DeepSeek, and DeepSeek deducts them from your prepaid balance. Conduit takes nothing.

**When is DeepSeek cheaper?**

Off-peak, when every price is half the peak price. DeepSeek's peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday, excluding Chinese public holidays; all other hours, weekends and those holidays are off-peak.

**Does DeepSeek's thinking show in Conduit?**

Yes. DeepSeek's models think before answering by default, and Conduit shows that reasoning above the answer, folded once the reply is done. Conduit has no switch to turn DeepSeek's thinking off.

**Can I use MCP tools with DeepSeek?**

Yes. DeepSeek's models call tools in thinking mode, and Conduit sends each round's reasoning back the way DeepSeek requires. A tool asks before it runs unless it is marked read-only.

**Where does DeepSeek store my data?**

DeepSeek's privacy policy says it collects, processes and stores personal data in the People's Republic of China. If that doesn't suit your work, Together AI and Fireworks AI serve DeepSeek models from their own infrastructure, and both are built into Conduit.

**Can I chat with my documents using DeepSeek?**

Not while DeepSeek is the active provider. DeepSeek has no model for indexing documents, so Conduit can't build a document collection with it. Switch to OpenAI, Gemini, OpenRouter, or Ollama to index your files.

Sources for the DeepSeek facts, checked 4 October 2026: DeepSeek's [models and pricing](https://api-docs.deepseek.com/quick_start/pricing), [context caching](https://api-docs.deepseek.com/guides/kv_cache), [thinking mode](https://api-docs.deepseek.com/guides/thinking_mode), [privacy policy](https://cdn.deepseek.com/policies/en-US/deepseek-privacy-policy.html) (last updated 10 February 2026), and [open platform terms](https://cdn.deepseek.com/policies/en-US/deepseek-open-platform-terms-of-service.html). Conduit rows match v1.0.0-rc.7 and are checkable in [the documentation](https://conduitllm.com/docs.html) and [the source](https://github.com/runningpixels/conduit). Found an error? [Tell us](https://github.com/runningpixels/conduit/discussions) and we will fix it.

## Related

- [Use OpenRouter in a desktop app](https://conduitllm.com/providers/openrouter.html) — Use OpenRouter in a free, open-source desktop app. Setup in four steps, what it costs, free-model limits, web search, and what doesn't work yet.