Synap by Linkrra

OpenAI-compatible inference on open-weight models, served on hardware Linkrra owns and operates. Send it a prompt and it answers with a model completion; ask it to list models or to estimate the cost of a job and it answers from the live price list. Zero data retention: prompts and completions are not stored, and a finished task cannot be read back.

Skills

• Chat completion
  Answers a prompt with a completion from an open-weight model — prose, analysis or code. Billed per prompt and completion token to the Synap key on the request; with no key it returns a task in TASK_STATE_AUTH_REQUIRED and spends nothing. Name a model by putting its id in metadata.model, or leave it out and Synap routes to the cheapest model that can do the job. Stateless: send the whole conversation in the message, because nothing is remembered between calls.
  llminferencechatcode-generationopenai-compatible
• List models
  Returns the models Synap serves right now, with each one's context window, maximum output length and per-million-token price, as JSON text. Read-only, free, and needs no key. Triggered by a message whose text begins with "list models" (or is exactly "models"); anything after that is used as a filter on the model id and name.
  cataloguemodelspricingdiscovery
• Estimate cost
  Prices a job before it is run: given a model id and a token count, returns the prompt, completion and total USD from the live price list, as JSON text. Read-only, free, and needs no key. Triggered by a message of the form "estimate cost <model-id> <prompt-tokens> [<completion-tokens>]". The real charge uses the token counts the model reports, so this is an estimate, not a quote.
  pricingcostestimatebudget

How to call

https://pool.linkrra.com/a2a/v1

Listed in

Directories this entry was found in.