---
name: lingtai-ai/preset-skill-deepseek
source: https://app.decimal.ai/s/lingtai-ai-preset-skill-deepseek@3/SKILL.md
source_sha256: faa240e9064a
---

# `deepseek`

`deepseekPreset()` in `tui/internal/preset/preset.go` uses the
shared OpenAI-compatible text shape: provider `deepseek`, default model
`deepseek-v4-pro`, `https://api.deepseek.com`, and `DEEPSEEK_API_KEY`.
The shipped manifest has no `vision` capability, so this manual records no
direct image route and no DeepSeek plan-level vision MCP.

## Template-specific settings

Read the official [DeepSeek API introduction](https://api-docs.deepseek.com/)
when checking current models, protocol, or limits. For an image request,
report that this preset's shipped wiring is text-only and let the agent
discover an explicitly chosen skill if one exists; do not guess credentials,
switch providers, or auto-load/invoke an MCP. Verify the
provider/model/endpoint/env-var fields in TUI source after a template change.

The deepseek preset's `base_url` offers three options in the TUI preset
editor (`ProviderRegionURLs["deepseek"]`):

- **DeepSeek API** — `https://api.deepseek.com`, `DEEPSEEK_API_KEY`
  (the template default).
- **OpenCode Go** — `https://opencode.ai/zen/go/v1`, `OPENCODE_GO_API_KEY`.
  This option is scoped to **DeepSeek models served through the OpenCode Go
  subscription**: the preset keeps `provider: deepseek` and the model row
  stays on the DeepSeek lineup (`deepseek-v4-pro` / `deepseek-v4-flash`,
  both served by `/zen/go/v1`). The `zhipu`, `minimax`, `mimo`, `kimi`, and
  `grok` presets carry their own **OpenCode Go** `base_url` row (same URL,
  same `OPENCODE_GO_API_KEY` slot) — route GLM, MiniMax, MiMo, Kimi, and
  Grok through those, not through this one. For the remaining Go models
  (gpt-5.6-luna, qwen, ...) create a **Custom** preset instead, where
  the model row is free-text and `wire_api` stays selectable. See the
  official [OpenCode Go docs](https://opencode.ai/docs/go/). China-hosted
  `deepseek-v4-flash` requires opting in at the OpenCode workspace page
  before first use.
- **Custom** — free-typed endpoint. The editor leaves `api_key_env` alone on
  this row, and a saved edited built-in gets `stampAutoEnvVar`'s numbered
  slot. To change the slot name itself, edit the preset JSON.

Selecting the DeepSeek API or OpenCode Go option also switches `api_key_env`
to the matching slot in the editor. Saving an **edited built-in** keeps that
slot only for OpenCode Go, whose credential is a shared cross-provider account
(the same `OPENCODE_GO_API_KEY` the `zhipu` and `minimax` presets use). On the
DeepSeek API row the saved preset gets its own numbered slot instead —
`DEEPSEEK_1_API_KEY`, `DEEPSEEK_2_API_KEY`, ... — because `DEEPSEEK_API_KEY` is
the template's shared slot, and keeping it would let a second deepseek preset
overwrite the first one's key.

## Operations

For base URL/API-compat/model/capability declaration shape versus
credentials, see `reference/operations/endpoint-capabilities/SKILL.md`.