Claude Code glossary for teams

The word means two things. A subscription seat is one person's Claude or ChatGPT plan, such as Claude Max or ChatGPT Business: a flat monthly fee, paid whether it is used or not, and never billed per token. A Selan seat is a person in your workspace or a CI token.

More: Claude Max plan vs API cost

Metered billing has no monthly fee. It bills every token, at a rate set by the model and by the kind of token. An API key from Anthropic, OpenAI or OpenRouter is billed this way.

More: where the break-even sits between a plan and the API

An API key issued by a provider such as Anthropic or OpenAI, billed per token on the account it belongs to. On a laptop it has left your control: it can be copied into a .env file, committed with a repository, or read by any program started from that shell.

More: Claude Code without API keys on laptops

A tool that runs on the provider accounts and keys you already hold, while the provider bills you directly at its own prices. Selan works this way. It holds no provider account of its own to resell to you.

More: Providers and models on your own accounts

Usage credits let a Claude plan keep working past its usage limit. They are billed at standard API rates, up to a monthly cap you set. On Team and Enterprise the request goes to an admin.

More: the ways past a Claude Code usage limit

A gateway sits between the coding tools and the providers, so every request passes through one place that knows who sent it. It holds the provider keys and attaches one to each request on its way out. Selan is a hosted gateway that a team signs in to.

More: How Selan works, and LiteLLM and Selan compared

On a Claude subscription, the cap on how much you can use in a window. There are two, a session limit and a weekly limit, and every request counts against both. Claude Code, claude.ai and Claude Desktop share one allowance.

More: Claude Code rate limit reached: resets and fixes

The usage limit that resets every five hours. When it is full, Claude Code stops with You've hit your session limit and the time it resets.

More: what the limit message looks like

The usage limit across all models that resets once a week, at a fixed day and time assigned to your account. One heavy burst, such as a large fan-out of agents, can use it up before the session limit resets.

More: when each limit resets

A cap on how fast one credential may send requests. On an API key it is a per-minute limit, and a request over it comes back as API Error: Request rejected (429). People also say rate limit for a plan's usage limit, which is a different cap.

More: the rate limit messages, one by one

A Selan cap on what is billed per token, set per key or per person, for a day, a week or a month in UTC. It binds the moment it is reached, and the refusal names the limit and when it lifts. A person's own limit leaves the company's subscription seats serving them.

More: what happens at a spend limit, and spend limits per developer

A session's working memory: the conversation, file contents, command output, CLAUDE.md and system instructions. Its size is set by the model. Through Selan, a model named with --model gets its real window, so a 262k model compacts at 262k and not at an assumed 200k.

More: running any model in Claude Code, and the ctx meter on the statusline

What Claude Code does when the context window nears its limit: it clears older tool output, then summarises the conversation to make room. /compact does it on demand.

More: which window a picked model compacts against

A token is a piece of text, such as a word or part of one, and providers price them per million. Input is what a request sends, and output is what the model writes back. Output costs five times as much as input on Claude Opus 5.5 and Sonnet 5.5.

More: the API rates, model by model

Every request carries the conversation so far. The prompt cache keeps the start of a request for five minutes, or an hour at a higher price, so a later request that begins the same way reads it back instead of paying to process it again.

More: cache reuse per repository

Storing the start of a request in the cache, the first time it is sent. A five-minute cache write costs 1.25 times fresh input. A one-hour cache write costs twice fresh input.

More: cache rates beside input and output

Reading a cached start back on a later request. On Claude Opus 5.5 and Sonnet 5.5 it costs a twentieth of fresh input, so a session that reuses its context costs far less than its token count suggests.

More: usage per developer, where new tokens leave cache reads out

A file at the root of a repository that Claude Code sends with every request: the commands, rules and traps every session needs. Its size times the number of requests is often the largest single cost.

More: what a repository's sessions cost, and what to cut

A program that gives Claude tools and data, such as an issue tracker or a database, over the Model Context Protocol, an open standard. Claude Code starts one from a command or reaches one at a URL. With Selan, an owner switches a connector on and each developer signs in to it as themselves.

More: every MCP server Selan connects Claude to

A folder with a SKILL.md file of instructions, and any scripts it needs, that Claude loads when a task calls for it or when you type /skill-name. Its body loads only when it is used, so a long skill costs almost nothing until then.

More: the top Claude Code skills

A helper Claude Code starts for a side task, in its own context window, with its own system prompt and tools. It returns a summary to the main conversation. Its requests count toward the same usage limits as the main conversation.

More: Anthropic's subagent docs

A bundle of skills, subagents, hooks and MCP servers that Claude Code installs and loads as one unit. Most plugins come from a marketplace.

More: selan-practices, a plugin that audits a repository's sessions

A catalog of plugins: a repository with a .claude-plugin/marketplace.json file that lists each plugin and where to fetch it. In Selan, an owner switches a marketplace on and its plugins appear for every developer's Claude. Nothing installs on its own.

More: How Selan works

A Selan credential an owner mints for a job with no browser, such as CI, a container or cron, carried in SELAN_TOKEN. It spends as itself, under an address of its own, so the job's bill shows as the job. A CI token counts as a Selan seat.

More: running Claude Code in CI, containers or cron

The list /model shows in Claude Code. With Selan it is your company's: every model an owner has switched on and put in the picker, Claude models first. A model left out can still be named with --model.

More: which models are in the picker

Selan replaces a recognised secret in a request with a placeholder, such as [REDACTED: jwt], before the request reaches the provider. It redacts and forwards, and never blocks a request over a secret.

More: Claude Code secret scanning and redaction

The line at the bottom of Claude Code. Selan's shows the company's 5-hour and 7-day headroom and your own spend against your own budget.

More: what each meter means