Accounts and access
| Term | Meaning | Applies to |
|---|---|---|
| Admin | A workspace role that manages admin-only settings, such as budgets, spend rules, guardrails and trust policies, and reads the audit log. | API, Dashboard, CLI, MCP |
| Member | A workspace role that uses the workspace and can see its spend rules, but can’t change owner- and admin-only settings. | API, Dashboard, CLI, MCP |
| Owner | A workspace role that manages the same admin-only settings as an admin and receives owner emails, such as model deprecation alerts. | API, Dashboard, CLI, MCP |
| Workspace | A container that keeps its own API keys, routing settings, presets, guardrails and admin access separate from other workspaces. | API, Dashboard, CLI |
Keys and authentication
| Term | Meaning | Applies to |
|---|---|---|
| API key | A secret starting phaseo_v1_sk_, sent as a bearer token to authenticate model and generation requests. | API, Dashboard, CLI, SDK, MCP |
| BYOK | Bring your own key: Phaseo routes requests with your provider API key, and the provider bills the model usage. | API, Dashboard |
| BYOK fallback | A workspace setting that decides whether a failed BYOK request may retry on Phaseo credits. | API, Dashboard |
| Capability | A permission on a management key or OAuth sign-in, such as keys:write, that decides which administration APIs it can call. | API, Dashboard, CLI, MCP |
| Management key | A secret starting phaseo_v1_mk_ for administration APIs such as keys, presets and guardrails; it can’t send inference requests. | API, Dashboard |
| OAuth app | An app that signs people in with Phaseo OAuth and acts for them within the scopes and workspace they approve. | API, Dashboard, CLI, MCP |
| Trusted workload | A deployment or CI job that sends its platform’s short-lived OIDC token instead of an API key, matched by a trust policy. | API, Dashboard |
Models and providers
| Term | Meaning | Applies to |
|---|---|---|
| Context length | The context size a provider route supports for a model, shown for each provider on the model page. | API, Dashboard |
| Deprecated (model) | The lab or provider has announced the model’s retirement, and its requests are still served as usual. | API, Dashboard |
| Model ID | The ID you send as model, such as openai/gpt-6-astra, shown in the quickstart on each model page. | API, Dashboard, CLI, SDK, MCP |
| Private model | A dedicated or self-hosted OpenAI-compatible deployment added to your workspace, which only your workspace can see or call. | API, Dashboard |
| Provider | A company that hosts a model and sets its availability, supported parameters, pricing and performance. | API, Dashboard, SDK, MCP |
| Provider status icon | The status Phaseo sets by hand for a provider route: Active, Lower priority (backup only) or Disabled. | Dashboard |
| Provider-qualified model ID | A model ID with a provider prefix, such as baseten:thinking-machines/inkling-small, that runs on that provider with no fallback. | API, SDK |
| Quantization | Storing a model’s weights in a smaller number type, such as FP8 or INT4, which serves faster at some cost to quality. | API, Dashboard, SDK |
| Retired (model) | The retirement date has passed; requests work while a provider serves the model, then fail with 410 model_retired. | API, Dashboard |
| Routable model | A model with at least one provider route active on the Gateway now; GET /v1/models lists only these by default. | API, Dashboard, MCP |
Requests and responses
| Term | Meaning | Applies to |
|---|---|---|
| Cached tokens | Input tokens the provider read from its prompt cache, billed at the cache-read rate and reported as cached_tokens. | API, SDK |
| Gateway | The Phaseo API at https://api.phaseo.app/v1, which routes each request to a provider that hosts the model. | API, SDK |
| Generation ID | The G- request ID returned in x-request-id and error bodies; use it for logs, lookups and support. | API, Dashboard, SDK |
| Prompt caching | Keeping the start of a prompt the same so the provider reads it from cache and bills it at the cache-read rate. | API, SDK |
| Reasoning effort | How much a reasoning model thinks before answering, set with reasoning.effort or reasoning_effort, from none to max. | API, SDK |
| Reasoning tokens | Tokens a reasoning model spends before it answers; they count as output and use part of the output limit. | API, SDK |
| Service tier | A speed and price tier for a text request (standard, fast, ultrafast or flex), served only by providers offering it. | API, SDK |
| Session ID | A session_id you send to keep a conversation’s requests on one provider, for up to 24 hours of activity. | API, Dashboard, SDK |
| Streaming | Setting stream: true so the Gateway sends output as server-sent events while the model writes it. | API, SDK |
| Tokens | The units a model reads as input and writes as output; every token adds latency and cost. | API, Dashboard, SDK |
| Usage object | The usage field in a response, which reports input, output and cache token counts for the request. | API, SDK |
Routing
| Term | Meaning | Applies to |
|---|---|---|
| Alpha and beta provider channels | Routing settings toggles so preview providers get traffic only when you opt in; provider.include_alpha allows alpha providers per request. | API, Dashboard |
| Auto Routing | Send phaseo/auto and Phaseo picks a model for each request from the workload, your objective and price limits. | API, Dashboard, CLI, SDK, Slack |
| Dynamic route | A versioned flow of rules attached to API keys that can branch, split traffic, cap usage and switch models. | API, Dashboard |
| Free router | Send phaseo/free and Phaseo runs the request on an available free model, with no credits needed. | API, SDK |
| Model fallbacks | Models listed in models, tried in order when the previous model ends in 429, 500, 502, 503 or 504. | API, SDK |
| Preset | A saved model, system prompt, parameters and provider rules that callers use by sending its name, such as @support-triage. | API, Dashboard, SDK |
| Preset version label | A name such as production that points at one published preset version; moving it rolls requests forward or back. | API, Dashboard, SDK |
| Prompt template | Preset messages with {{variables}}, so each request sends only the preset reference and the variable values. | API, Dashboard, SDK |
| Provider fallback | When the chosen provider fails, the Gateway tries the next eligible provider for the same model. | API, SDK |
| Provider sorting | How providers are ranked: balanced (default, cost per successful answer), price, latency or throughput. | API, Dashboard, CLI, SDK |
| Regional routing | Sending text requests to eu.api.phaseo.app or us.api.phaseo.app to use only routes that run there. | API, SDK |
| Retired model routing | An opt-in workspace setting that sends requests for a model no provider serves to an approved or chosen replacement. | API, Dashboard |
Guardrails and privacy
| Term | Meaning | Applies to |
|---|---|---|
| Guardrail | A policy on API keys or members that limits providers, models and data handling, and scans prompts; it doesn’t cap spend. | API, Dashboard |
| Guardrail action | What a guardrail content check does when it matches: Flag, Redact or Block. | API, Dashboard |
| I/O logging | A privacy setting that stores each request’s prompt and response, so you can view, replay or open it in Chat. | API, Dashboard, Chat |
| Prompt injection check | A guardrail check that scans user-supplied content for common prompt injection patterns. | API, Dashboard |
| Sensitive info check | A guardrail check that finds data such as email addresses, phone numbers and card numbers before the request leaves Phaseo. | API, Dashboard |
| ZDR | Zero data retention: the provider doesn’t keep prompts or outputs; require it per request, per guardrail or per workspace. | API, Dashboard, CLI, SDK, MCP |
Tools and outputs
| Term | Meaning | Applies to |
|---|---|---|
| Decisions API | POST /v1/decisions, which answers typed questions (predicate, choice, score) with probabilities, in the order asked. | API, SDK |
| Response healing | A plugin that repairs almost-valid JSON, such as JSON in a Markdown fence, before it reaches your parser. | API, SDK |
| Server tool | A phaseo: tool Phaseo runs when the model calls it, such as phaseo:web_search, phaseo:web_fetch, phaseo:datetime or phaseo:advisor. | API, SDK, Chat |
| Structured outputs | Sending a JSON Schema so the model returns JSON in that shape instead of free text. | API, SDK |
| Tool calling | Describing your functions in tools so the model returns a call your app runs, then answers with the result. | API, SDK |
Cost, limits and billing
| Term | Meaning | Applies to |
|---|---|---|
| Budget | A workspace spend ceiling (daily, weekly, monthly or lifetime); requests past it get 429 key_limit_exceeded. | API, Dashboard |
| Credits | Prepaid workspace balance that paid requests spend at catalog prices; purchases carry a 5% fee ($1 minimum), and :free models need none. | API, Dashboard, CLI, MCP |
| Department budget | A daily, weekly or monthly spend ceiling on the keys and OAuth requests that count toward one department. | API, Dashboard |
| Key limit | A daily, weekly or monthly request or spend cap on one API key, checked on every request. | API, Dashboard, CLI |
| Rate limit | The requests-per-minute allowance for each user in each workspace, which rises as the workspace becomes established. | API, CLI |
| Spend alert | The notification a spend rule sends when spend, a forecast, a spike or the balance crosses its threshold. | API, Dashboard |
| Spend rule | A rule that alerts when spend for a scope crosses a threshold, and can block further requests with 429 spend_rule_blocked. | API, Dashboard |
Observability
| Term | Meaning | Applies to |
|---|---|---|
| App attribution | Headers such as X-App-Id, X-Title and HTTP-Referer that group a request’s usage under one app. | API, SDK |
| Audit log | The record of who changed keys, members and other admin controls, which only owners and admins can read. | API, Dashboard, CLI |
| Broadcast | Sends a trace of each gateway request to an observability tool, webhook, storage bucket or data warehouse, with no code changes. | API, Dashboard |
| Open in Chat | Continues a stored request or session turn as a conversation in Chat, with the same model or another. | Dashboard, Chat |
| Request classifier | A taxonomy and a model that label each finished request, stored as request labels you can filter by. | API, Dashboard |
| Request label | A key/value pair, such as team=support, sent in the X-Phaseo-Metadata header to filter spend and logs. | API, SDK |
| Request logs | A record for each request with its usage, latency, cost, errors and the provider that served it. | API, Dashboard, CLI, MCP |
Async jobs
| Term | Meaning | Applies to |
|---|---|---|
| Async job | A video or batch request that returns a job at once; you poll it or wait for a webhook until it finishes. | API, SDK |
| Batch | A Batch API job for large deferred workloads, where each row targets one model. | API, SDK |
| Video job | A Video API request that returns a job; download the video once its status is completed. | API, SDK |
| Webhook endpoint | A workspace URL that receives signed status events for video and batch jobs. | API, Dashboard |
Enterprise
| Term | Meaning | Applies to |
|---|---|---|
| Department | A group of workspace members, managed by hand or synced from your directory, that budgets, rate limits and spend rules can target. | API, Dashboard |
| Directory | The workspace’s enterprise departments and each member’s effective entitlements, managed by owners and admins. | API, Dashboard |
| Identity add-on | The workspace add-on needed to turn on SSO or SCIM, create provisioning tokens and manage the directory. | API, Dashboard |
| SCIM | Provisioning workspace members and departments from your identity provider with a SCIM provisioning token. | API, Dashboard |
| Self Serve Enterprise | A monthly workspace subscription for SSO, SCIM, directory management, departments and priority email support. | Dashboard |
| SSO | Single sign-on for workspace members through SAML or a custom OIDC provider. | API, Dashboard |
CLI, MCP, SDKs and apps
| Term | Meaning | Applies to |
|---|---|---|
| Agent skills | Public Phaseo instructions for coding agents, covering the Gateway, SDKs, presets, routing and migrations. | API, CLI, SDK |
| Client SDK | A Phaseo package for your language with one method per Gateway endpoint, such as @phaseo/sdk or phaseo. | SDK |
| Docs MCP | A read-only MCP server at https://phaseo.app/docs/mcp that lets coding assistants search and read these docs. | MCP |
| Phaseo CLI | The phaseo command for creating keys, changing settings and automating admin work from a shell or CI. | CLI |
| Phaseo MCP | An MCP server that gives AI clients live model, pricing, usage and workspace data, and can optionally manage your account. | MCP |
| Slack agent | The Phaseo agent in Slack, which answers mentions and direct messages with the model your workspace chose. | Dashboard, Slack |
Feature stages
| Term | Meaning | Applies to |
|---|---|---|
| Alpha | An early feature that may change a lot or be removed. | API, Dashboard, CLI, SDK, MCP, Chat, Slack |
| Beta | A feature that works for its main use, though details may change; changes are announced in the changelog. | API, Dashboard, CLI, SDK, MCP, Chat, Slack |
| Deprecated (feature) | A feature that still works but will be removed; its page gives the removal date when one is set. | API, Dashboard, CLI, SDK, MCP, Chat, Slack |
| Feature Preview | The settings page that shows whether the Video and Batch APIs are turned on for your workspace. | Dashboard |
| On request | A feature that is turned on per workspace; until it is, its settings page is hidden and its API returns an error. | API, Dashboard |
| Generally available | A stable feature, shown with no stage badge. | API, Dashboard, CLI, SDK, MCP, Chat, Slack |