API reference

77th Junction is OpenAI-compatible. Anything that works with the OpenAI API works here after you change the base URL and the key.

Quick start

Base URL

https://77thjunction.app/v1

curl https://77thjunction.app/v1/chat/completions \
  -H "Authorization: Bearer sk-77j-YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "junction-forge",
    "messages": [
      {"role": "user", "content": "Write a haiku about midnight code"}
    ]
  }'

Authentication

Send your key in the Authorization header. Create and revoke keys in the dashboard. Keep keys secret: anyone with a key can spend your balance.

Authorization: Bearer sk-77j-…

Models

GET https://77thjunction.app/v1/models lists the models your key can use.

Model ID	Context	Input / 1M	Output / 1M
`junction-swift` Fast · reads images · 262K context	256K	$1.50	$4.50
`junction-forge` Coding & agent · 1M context	1M	$4.50	$7.50
`junction-titan` Newest large model · 1M context	1M	$4.50	$7.50

Chat completions

POST https://77thjunction.app/v1/chat/completions, the same body as OpenAI: model, messages, stream, tools, temperature, max_tokens, response_format.

With stream: true you get server-sent events. Add "stream_options": {"include_usage": true} to receive the token counts in the last chunk.

Some models think before answering; that thinking streams as delta.reasoning and is billed as output.

Billing

Each request is charged after it finishes, for the exact tokens the model reports. Cached input tokens cost less. Failed requests are not charged. Requests are refused with 402 once the balance drops below one cent. You can set a spend limit on each key.

Errors

Errors use the OpenAI shape: {"error": {"message", "type", "code"}}.

Status	Code	Meaning
401	invalid_api_key	Missing, wrong or revoked key.
402	insufficient_quota	Balance is empty. Top up in Billing.
402	key_limit_reached	This key hit the spend limit you set on it.
403	account_suspended	The account was suspended.
404	model_not_found	No enabled model with that name.
429	rate_limit_exceeded	More than 120 requests a minute on one key.
502	upstream_error	The model failed. Nothing was charged.