Cherry Studio

Cherry Studio is a free, open-source desktop chat client for Windows, macOS, and Linux. It calls models through provider “types”, and every type can point at MirAPI: an OpenAI-type provider is the primary path, while the Anthropic and Gemini types are there when you need those protocols’ semantics.

What you need

Element	Value
Base URL (OpenAI)	`https://api.mirapi.ai/v1`
Base URL (Anthropic)	`https://api.mirapi.ai`
Base URL (Gemini)	`https://api.mirapi.ai/v1beta`
API key	Your MirAPI key (`sk-...`)
Model	`deepseek-chat` (from `GET /v1/models`)

The same key works across all three provider types and every assistant in the app.

Configure

Open Settings → Model Services → Add.
Choose the OpenAI provider type.
Fill in:

Provider name: MirAPI
API base URL: https://api.mirapi.ai/v1
API key: sk-...

Add a model. In the provider’s model list, click Manage / Add model and enter the exact id from the catalogue:

deepseek-chat

Save and enable the provider.

The key is sent as Authorization: Bearer; the scheme is case-insensitive and a bare key without the scheme is also accepted. Cherry Studio appends /chat/completions to the base URL, which is why the address must include /v1.

Verify with a message

Start a new assistant using the MirAPI provider and send:

Reply with exactly: connected to MirAPI

Expect the exact phrase in the reply. If the call errors, keep the message Cherry Studio shows — it ends with the MirAPI request id, which support needs.

Protocol variants

Cherry Studio also ships Anthropic and Gemini provider types. Both talk to MirAPI through the matching endpoints; pick one when you specifically want that protocol’s semantics — for example Anthropic’s cache_control prompt caching.

Anthropic

Element	Value
Base URL	`https://api.mirapi.ai` (no `/v1`)
API key	Your MirAPI key (`sk-...`), sent as `x-api-key`

The Anthropic type calls /v1/messages. MirAPI accepts the key as either x-api-key or Authorization: Bearer. Enter the bare base URL — the client appends /v1/messages itself, so do not add /v1.

Gemini

Element	Value
Base URL	`https://api.mirapi.ai/v1beta`
API key	Your MirAPI key (`sk-...`), sent as `x-goog-api-key`

The Gemini type calls generateContent on /v1beta/models/{model}:generateContent. MirAPI accepts x-goog-api-key or a ?key= query parameter; the bare https://api.mirapi.ai is also accepted as a base URL.

Billing

Chat is billed per token: USD per 1M tokens, with separate input, output, and cache-read tiers. Reasoning tokens are billed as output, and cache reads cost less than plain input.
MirAPI uses a prepaid USD balance — no subscription, no minimum, and the balance never expires.
When the balance runs out, calls return 403. Top up and the same key works again immediately.
Reconcile spend in the console billing log, with GET /api/usage/token for the account summary and GET /api/log/token for per-request detail.

Troubleshooting

401 — the key is missing, malformed, or unknown. Check it and retry.
403 — the key is valid but the request was rejected: the model is not in the key’s whitelist, the balance is exhausted, the quota is reached, or the caller’s IP is not allowlisted. Fix the indicated condition and retry.
429 — rate limited. MirAPI sends no Retry-After header, so back off with jitter (1s → 2s → 4s, capped around 30s).
500 — an upstream failure; retry idempotent requests safely. MirAPI never silently swaps in another model, so the retry stays on the model you chose.
Add models by their exact GET /v1/models id — a typo is an error, not a fallback.