Glossary

Terms used in the Syntic Developer Platform documentation, in alphabetical order. Where a term names a field, header or error, it is shown in code style.

Admin API

A planned set of endpoints under /v1/organizations/* for managing members, workspaces, keys and reports with an admin key. In preview; see Admin API keys.

Allowance (plan allowance)

The number of tokens your subscription plan covers, enforced over monthly, weekly and 5-hour session windows. Requests inside the allowance carry no per-token charge. See Limits.

API key

A secret credential sent as Authorization: Bearer <key>. Private keys start with sk_live_ and may call the API; public keys (pk_live_) are refused with 403 public_key_rejected. See Authentication.

Assistant message

A message with "role": "assistant": a reply the model gave earlier, or one you write yourself as an example. It may carry tool_calls.

Billing source

Which funds paid for a request, reported in the x-syntic-billing-source header: subscription (plan allowance) or api_credit (usage credits).

Cached input (cached tokens)

Prompt tokens that match the beginning of an earlier request and are served from the prompt cache, billed at a lower rate. Reported as usage.cached_tokens. See Pricing.

Chat completions

The Syntic API’s main endpoint, POST https://app.syntic.ai/v1/chat/completions. It takes a list of messages and returns the assistant’s next message. See Chat completions.

Choice

One candidate reply inside choices. Syntic always returns exactly one, at index 0.

Chunk

One server-sent event in a streamed reply, with "object": "chat.completion.chunk". Text arrives in choices[0].delta.content.

Console

The web app at platform.syntic.ai where you manage keys, organizations, usage, billing and agents. See Console & admin.

Context window

The most tokens a single request can hold, prompt and reply together: 262,144 tokens for every Syntic model.

Default model

The model that answers when model is missing, unknown, or not included in your plan: Amara S1.2.

Downgrade (model fallback)

When you ask for a model your plan does not include, the default model answers instead and the response carries x-syntic-model-downgraded with the answering model’s name.

Finish reason

Why the reply ended, in choices[0].finish_reason: stop (complete), length (hit max_tokens or the context window) or tool_calls (waiting for your tool results). See Stop reasons.

Identity instruction

A short Syntic instruction the API places ahead of your system message. Your instructions follow it unchanged; questions about the underlying model are answered as Syntic AI.

Message

One entry in the messages array, with a role (system, user, assistant or tool) and content. A request may hold up to 500 messages.

Model name

The exact string that selects a model, such as Amara S1.2 or Syntic Code. Names include spaces and capitals and must match exactly. See Models overview.

Organization

The account that owns keys, members, plan, credits and usage. Each API key belongs to one organization, which pays for its requests. See Organizations.

Origin allowlist

A list of web origins attached to a key. When set, requests must carry a matching Origin (or Referer) header, or they are refused with 403 origin_not_allowed.

Preview

A feature whose design is published and being built. Pages marked Preview are the specification; the fields they describe are ignored or unavailable until release.

Prompt

Everything the model reads before replying: system message, conversation, documents, tool definitions and tool results. Measured in usage.prompt_tokens.

Prompt caching

Automatic reuse of an identical prompt beginning from an earlier request. Cached tokens are billed at the cached-input rate. Keep stable content first to benefit.

Quota exhausted

The 402 error returned when neither the plan allowance nor usage credits can pay for a request. Its limit field says why. See Handle 402.

Rate limit

The cap on how many requests a key may send per minute: 600, over a sliding 60-second window. Exceeding it returns 429 with Retry-After.

Request ID

A unique identifier for each request, in the x-syntic-request-id response header. Quote it when you contact support. See Request IDs.

Server-sent events (SSE)

The streaming format: a sequence of data: lines over one HTTP response, ending with data: [DONE]. See Stream format.

Spend limit

A monthly cap on usage-credit spending, set by administrators for the organization, a group or a member. Reaching it returns 402 with limit: "spend_limit".

Streaming

Receiving the reply piece by piece as it is generated. On by default; set "stream": false for a single JSON object.

System message

The first message, with "role": "system", holding instructions for the whole conversation. See Use the system message.

Token

The unit models read and write, roughly three to four characters of English text. Usage, limits and prices are counted in tokens.

Tool (function)

A function you describe in tools that the model may ask your code to run. The model returns tool_calls; you run them and reply with tool messages. See Tool calling.

Tool call

A request from the model to run one of your tools, with an id, the function name and arguments as a JSON string.

Tool loop

Code that alternates between calling the model and running the tools it asks for, until a reply arrives with no tool calls. See Write a tool loop.

Usage

Token counts for a request, in the usage object: prompt_tokens, completion_tokens, total_tokens, and cached_tokens when part of the prompt was cached.

Usage credits

A prepaid dollar balance that pays for requests beyond the plan allowance, at each model’s per-token rate. See Billing & credits.

Zero data retention (ZDR)

An organization mode in which models that need standard data retention are unavailable; requests for them return 400 model_unavailable.