API reference

All endpoints below are relative to https://arzarouter.arzastore.com/api — that is, https://arzarouter.arzastore.com/api/v1/….

Authentication

Send your key as a bearer token on every authenticated request. Missing or malformed headers get 401 with an OpenAI-shaped error body.

Header

all authenticated endpoints

Authorization: Bearer sk-infer-...
Content-Type: application/json

Account and inference

Method	Path	Purpose
POST	/v1/chat/completions	OpenAI-compatible completion; supports `stream` and `tools`/`tool_calls`
POST	/v1/responses	OpenAI Responses API — `input` in, `output` array out; translated onto the same upstream call
POST	/v1/messages	Anthropic-compatible messages API
GET	/v1/models	Model IDs this key may call
GET	/v1/me	Account profile and balance
GET	/v1/balance	Balance, mirrored from the router meter
GET	/v1/usage	Recent usage records
GET	/v1/usage/by-model	Usage grouped by model
POST	/v1/keys	Create an API key (returned once)
GET	/v1/keys	List keys by prefix
DELETE	/v1/keys/:id	Revoke a key
POST	/v1/providers	Attach an upstream account
GET	/v1/providers	List attached upstreams
DELETE	/v1/providers/:id	Remove an upstream
POST	/v1/deposits	Register a deposit
GET	/v1/deposits/address	Your unique Solana deposit address
GET	/v1/deposits	Deposit history
GET	/v1/budgets	Your spend caps ($/Mtok), with the listed price of each target
PUT	/v1/budgets	Create or update a cap (`model`, `api_key_id`, `max_input_price`, `max_output_price`)
DELETE	/v1/budgets/:id	Remove a cap

MCP server

One sk-infer- key also authenticates an MCP server, so an agent can read and change account state without bespoke HTTP glue. Two transports: SSE (/mcp/sse then POST /mcp/messages?sessionId=…) and streamable HTTP (POST /mcp).

MCP — Claude Code / Cursor

SSE transport

claude mcp add --transport sse arzarouter \
  https://arzarouter.arzastore.com/mcp/sse \
  --header "Authorization: Bearer sk-infer-..."

# then ask: "list my arzarouter models" / "what is my balance?"

Tool	Purpose
`list_models`	Model ids with published $/Mtok pricing
`get_balance`	Credit units, USDC equivalent, deposit rate
`list_api_keys`	Keys by prefix (never the secret)
`create_api_key`	Mint a key — secret returned once
`list_budgets`	Spend caps by account, key or model
`set_budget`	Create or update a cap
`usage_by_model`	Token usage grouped by model
`list_deposits`	Deposit history with status and signature
`get_deposit_address`	Unique Solana USDC deposit address
`chat_completion`	Run one completion, billed and metered

Management API

The same resources are reachable under /api/<name>, matching the paths other ArzaRouter deployments document. One key authenticates both surfaces.

Method	Path	Purpose
GET	/api/keys	List keys by prefix
POST	/api/keys	Create a key (returned once)
GET	/api/me	Profile and balance
GET	/api/usage/logs	Recent usage records
GET	/api/catalog	Models with pricing, as this key sees them
GET	/api/market	Pool health counts and per-model availability
GET	/api/budgets	Your spend caps
PUT	/api/budgets	Create or update a cap
GET	/api/publisher/providers	Upstream accounts attached to your user
POST	/api/publisher/providers	Attach an encrypted upstream key
GET	/api/publisher/pool	Your attached capacity (earnings null — payouts not enabled)
GET	/api/deposit-address	Your unique Solana deposit address
GET	/api/topups	Deposit history
POST	/api/topups	Register a deposit

Public feeds

No key required. Every number is read from this deployment's own ledger; a metric we do not measure is returned as null with a note, never as a zero.

Method	Path	Purpose
GET	/api/market-ticker	SSE stream of per-model asks and last settled rate
GET	/api/pricing/cache	24h per-model prompt-cache window (cached tokens null)
GET	/api/pricing/usage	24h per-model requests and tokens
GET	/api/status/models	Per-model availability and latency (TTFT null)

Public (no key)

Method	Path	Purpose
GET	/v1/prices	Published per-Mtok prices for every sellable model (no auth)
GET	/v1/combos	Combo failover groups — call any of them as `combo/<slug>` (no auth)
GET	/v1/pricing	Published rate card (no auth)
GET	/v1/stats	Aggregate platform usage (no auth)
GET	/health	Liveness plus upstream reachability (no auth)

Serving model

ArzaRouter answers from its own upstream pool — it does not proxy another gateway. The catalog it can serve is what GET /v1/models returns, and each model carries the published price it is metered at. Requests are metered on ArzaRouter's own ledger: tokens in/out are weighted per model and deducted from your balance after each call.

A request that names a model outside the catalog is refused with 400 MODEL_UNKNOWN. Pool capacity is rotated automatically: a rate-limited or unhealthy credential is cooled down and the next one serves the call, so 429 from one upstream account is not surfaced to you unless every attempt fails.

Admin (pool operations)

These need the admin header x-inferhub-admin and are not part of the consumer API.

Method	Path	Purpose
GET	/admin/pool	Pool rows + health summary + catalog
POST	/admin/pool/import	Import credentials — `{"dir":"/path"}` or `{"tokens":"…"}`
POST	/admin/pool/probe	Health-probe one row (`{"id":N}`) or a batch
PUT	/admin/pool/models	Upsert a catalog entry (model, upstream id, cost weights, active)

Errors

400 malformed request or unknown model · 401 missing or invalid key · 402 insufficient balance, BID_EXCEEDED (priced above your cap), or PRICE_UNKNOWN (a cap is set for a model with no published price) · 404 unknown route · 502 upstream unreachable. Bodies are JSON with an error object carrying message and type.

Getting-started walkthrough: /docs.