@deep-tui/plugin-provider-deepseek

First-party DeepSeek provider for Deep TUI. It reads DEEPSEEK_API_KEY from the process environment when a request starts; secrets do not belong in deep-tui.config.json.

The model aliases map to DeepSeek's official API identifiers:

| Deep TUI model | DeepSeek API model | | --- | --- | | flash | deepseek-v4-flash | | pro | deepseek-v4-pro |

export DEEPSEEK_API_KEY="your-key"
deep-tui run --model flash "Quick task"
deep-tui run --model pro "Hard reasoning task"

The full API identifiers are also accepted.

Streaming is on by default, including split reasoning/text/tool-call deltas and final usage. Set streaming: false in this plugin's config to use the canonical non-streaming compatibility path.

Thinking mode is enabled with reasoningEffort: "max" by default. Set reasoningEffort: "high" to reduce it, or thinking: false to disable thinking and omit reasoning output.

Usage responses include prompt, cache-hit, cache-miss, output, and current context token counts. Both DeepSeek's native cache fields and the standard OpenAI-compatible cached-token detail field are recognized. The plugin calculates session charges from those actual billable token classes using the public DeepSeek per-million-token prices checked on 2026-08-21. Each request uses the tariff active when it starts:

| Period (UTC) | V4 Flash cache / input / output | V4 Pro cache / input / output | | --- | --- | --- | | Peak, 01:00–04:00 and 06:00–10:00 | $0.014 / $0.44 / $1.32 | $0.044 / $1.32 / $3.96 | | Off-peak, all other times | $0.007 / $0.22 / $0.66 | $0.022 / $0.66 / $1.98 |

Prices are USD per million tokens. The provider also implements the billing service using DeepSeek's live /user/balance endpoint. Override pricing when proxy or vendor rates differ; supplied fields are treated as fixed rates in both periods. The live account balance and provider invoice remain authoritative.