Skip to main content

Glossary

A

Acceptable use policy (AUP, acceptable use, prohibited use policy)

a list of uses a model licence forbids, usually incorporated into the licence by reference so that breaking it breaks the licence.

Open and closed models · See also Model licence

Agent instructions file (instructions file)

a markdown file in a repository that a harness reads into the context at the start of every session, holding build commands, conventions, and standing instructions.

Comparing harnesses · See also Harness

Agentic loop (agent loop)

the cycle of calling the model, running the tools it asks for, appending the results to the conversation, and calling again until the model asks for nothing.

The agentic loop · See also Harness, Turn, Stop condition

Audit log (audit logging, audit trail)

the record of which principal called which tool, with which arguments, against which server, and what came back. No single MCP server can produce one, because a server never learns which human is behind the client.

MCP gateways · See also MCP gateway, Trace

C

Cache prefix

the run of bytes from the start of a request that must match a stored request exactly for a cached prompt to be reused. The first difference ends the match.

Context management and compaction · See also Prompt caching

Capability negotiation

the declaration of which optional features each side supports, so neither a client nor a server uses one the other lacks.

MCP in depth · See also Model Context Protocol

Chain of thought (chain-of-thought, CoT)

written-out intermediate steps between a question and an answer. It improves accuracy on multi-step problems, and it is not a faithful record of how the model reached the answer.

Reasoning models · See also Reasoning model, Thinking tokens

Checkpointing (checkpoint, checkpoints)

a harness snapshotting the working tree as a run proceeds, so the files can be rolled back to an earlier point in the conversation.

Comparing harnesses · See also Harness

Chunking (chunk, chunks)

cutting documents into pieces before embedding them. The chunk size and the overlap between chunks decide what can ever be retrieved together.

Embeddings, vector stores, and RAG · See also Embedding, Retrieval-augmented generation

Closed model (closed models, proprietary model)

a model reachable only as a hosted service, whose weights are never released. The vendor can change or retire it under the name you call.

Open and closed models · See also Open-weights model

Cold start (cold starts)

the delay while a weights file is loaded into memory before the model can answer anything. It is the reason a model is usually kept resident.

Local and cloud inference · See also Inference server, Local inference

Compaction (compact, compacting)

replacing a stretch of conversation history with a generated summary of it, done by a harness at a threshold rather than at the point of failure.

Context management and compaction · See also Truncation, Context management

Context assembly

the harness's work of building the exact text sent on each call, from the system prompt and tool definitions through the history to the user's question.

What fills a context window · See also Context budget, Context management

Context budget

the plan for how many tokens each part of the context is allowed, and what gets dropped or summarised when the total will not fit.

What fills a context window · See also Context assembly, Compaction

Context management

deciding what stays in the context window, what leaves, and what replaces it, as a conversation grows past what the window holds.

Context management and compaction · See also Compaction, Truncation

Context window (context windows, window)

the maximum number of tokens a model can hold for one call. It is a single budget covering the whole request and the generated answer together, not an upload limit.

Tokens and the context window · See also Token, Input tokens, Output tokens

Continuous batching (batching, in-flight batching)

adding new requests to a batch that is already generating, as slots free up, so a card is rarely producing tokens for a single user.

Local and cloud inference · See also Inference server, Tokens per second

D

Decode (decoding)

the phase that generates output tokens one at a time, each one depending on the last. It is serial by construction and accounts for most of the wall-clock time of a long answer.

Inference and sampling · See also Prefill, Sampling

E

Embedding (embeddings, vector)

a fixed-length list of numbers standing for a piece of text, positioned by an embedding model so that similar meanings land near each other.

Embeddings, vector stores, and RAG · See also Vector store, Semantic search

F

Fallback chain (fallback, fallbacks)

an ordered list of models to try, where the next one gets the request when the one before it errors, refuses, or times out. A fallback to a weaker model succeeds quietly and changes the answer.

LLM gateways and routers · See also Model routing, Trace

Fine-tuning (fine-tune, fine tuning)

continuing to train a published model on your own examples so that it behaves differently. It changes the weights without revealing how they were made.

Open and closed models · See also Weights, Open-weights model

Foundation model (foundation models)

a model trained once, at great expense, on a broad corpus, then adapted or prompted for many different tasks rather than trained again for each one.

Start here · See also Large language model

Function calling

another name for tool calling, used by several vendors for the same mechanism of emitting a structured call for the caller to execute.

Tool calls · See also Tool call

G

Generative AI (generative artificial intelligence)

systems that produce new text, images, audio, or code from a prompt, rather than classifying or scoring something that already exists.

Start here

Greedy decoding (greedy sampling)

always taking the highest-probability token. It removes the sampler's randomness but not the provider's, so a hosted call is still not guaranteed to repeat.

Inference and sampling · See also Sampling, Temperature

H

Hallucination (hallucinations, hallucinate)

a fluent, confident output that is not true. The model produces likely-looking text and has no mechanism for telling what it does not know.

Start here

Harness (harnesses)

the program between you and the model API. It assembles the context, executes the tools the model asks for, enforces permissions, and keeps calling the model until the job is done.

The agentic loop · See also Agentic loop, Permission model

Headless mode (headless, print mode)

running a harness non-interactively. One prompt in, a result out, an exit code set, and nothing to confirm. It is what makes a harness usable in a pipeline.

Comparing harnesses · See also Harness, Stop condition

Hook (hooks)

a script the harness runs when an event fires, such as after a file edit, with no model involvement in the decision to run it.

Skills, plugins, and MCP · See also Plugin

combining keyword matching and vector search over the same corpus and merging the two result lists, so exact identifiers and paraphrases both work.

Embeddings, vector stores, and RAG · See also Semantic search, Reranking

I

Inference (inference call)

running a trained model to produce an output. The weights do not change and no learning happens.

Inference and sampling · See also Prefill, Decode

Inference server (serving stack, model server)

the software that keeps weights resident, exposes an API, manages the per-request caches, and batches many requests into one pass over the weights.

Local and cloud inference · See also Continuous batching, Tokens per second

Input tokens (input token, prompt tokens)

everything sent to the model in a request, including the system prompt, the tool definitions and every earlier turn of the conversation. Billed at a lower rate than output.

Tokens and the context window · See also Output tokens, Context window

K

Knowledge base

the documents a retrieval system searches, their embeddings, the store holding them, and whatever keeps the three in step.

Embeddings, vector stores, and RAG · See also Vector store, Retrieval-augmented generation

L

Large language model (large language models, LLM, LLMs)

a model trained to predict the next chunk of text, and big enough that predicting text well requires it to encode a great deal about the world.

Start here · See also Foundation model

LLM gateway (LLM gateways, AI gateway, model gateway)

a service inside an organisation that sits between applications and model vendors. It holds the vendor credentials and decides which model gets each call, what the call costs, and whether the caller is allowed to make it.

LLM gateways and routers · See also Model routing, Virtual key, MCP gateway

Local inference (local, self-hosted inference, on-premise inference)

running a model on hardware you control, so that no prompt or document leaves your network.

Local and cloud inference · See also Inference server, Quantization

Lost in the middle

the tendency of a model to use material placed in the middle of a long context less reliably than the same material placed at the start or the end.

What fills a context window · See also Context budget, Retrieval-augmented generation

M

MCP client

a connector inside the host that speaks the protocol to exactly one server. A host with three servers holds three clients.

MCP in depth · See also MCP host, MCP server

MCP gateway (MCP gateways)

a service that presents many MCP servers to a client as one. It holds the downstream credentials, decides which servers and tools each client may reach, and records what was done.

MCP gateways · See also LLM gateway, Policy enforcement, Audit log

MCP host (host)

the application the user interacts with, which holds the model connection, owns the conversation, and decides what is allowed to run.

MCP in depth · See also MCP client

MCP prompt (MCP prompts)

a template a server publishes for the user to invoke, often surfaced by a host as a slash command.

MCP in depth · See also Slash command, MCP resource

MCP resource (MCP resources, resource)

data a server exposes at a URI for the client to read. The application decides what to attach, rather than the model fetching it.

MCP in depth · See also MCP prompt

MCP server (MCP servers)

a separate program exposing tools, resources, and prompts over the protocol. It never sees the conversation and never talks to the model.

MCP in depth · See also stdio transport, Streamable HTTP transport

Memory

text written outside the model and put back into the context on a later call, so that something appears to carry over. The model itself retains nothing between calls.

Memory · See also Session memory, Persistent memory

Memory write

an update to a memory store during a session, decided by the model, by the user, or by a rule in the application.

Memory · See also Persistent memory, Recall

Message (messages)

one entry in the conversation history, carrying a role such as user, assistant or tool result, and its content. The whole list is resent on every call.

What fills a context window · See also Session memory, Context assembly

Model Context Protocol (MCP)

an open protocol, using JSON-RPC, for connecting AI applications to external tools and data, so each integration is written once rather than once per application.

MCP in depth · See also MCP host, MCP client, MCP server

Model coupling

how tightly a harness is bound to one vendor's models, and whether you can point it at another endpoint. It decides whether an LLM gateway can sit in front of it.

Comparing harnesses · See also Harness, LLM gateway

Model licence (model license, community licence)

the terms attached to a weight release. Often bespoke rather than Apache-2.0 or MIT, and often carrying conditions an open source licence could not.

Open and closed models · See also Acceptable use policy, Open-weights model

Model routing (routing, router)

choosing which model handles a request. The choice can be made by alias, by task class, by cost, by observed latency, or by which models are currently healthy.

LLM gateways and routers · See also LLM gateway, Fallback chain

N

Next-token distribution (next token distribution, logits)

the probability a model assigns to every token in its vocabulary as the next one, produced by applying softmax to the raw scores from one forward pass.

Inference and sampling · See also Sampling, Temperature, Top-p

O

Open-weights model (open weights, open model)

a model whose weights are published for download, under whatever licence the publisher chose. Downloadable does not mean open source.

Open and closed models · See also Weights, Model licence, Closed model

Output tokens (output token, completion tokens)

everything the model generates, produced one token at a time and billed at a higher rate than input, commonly three to five times higher. Thinking tokens count here too.

Tokens and the context window · See also Input tokens, Thinking tokens, Decode

P

Parallel tool calls (parallel tool use, parallel function calling)

several tool-call blocks in one reply, which the caller may run concurrently. Every one of them needs a matching result.

Tool calls · See also Tool call

Permission model (permission modes)

the rules a harness applies to decide which tool calls run without asking, which stop and wait for your confirmation, and which are refused outright.

The agentic loop · See also Sandbox, Harness

Persistent memory

a small store that survives the end of a session and is read back into the context at the start of the next one.

Memory · See also Recall, Memory write

Plugin (plugins)

a distributable, versioned bundle that can hold skills, slash commands, subagent definitions, hooks, and MCP server configuration together.

Skills, plugins, and MCP · See also Skill, Hook, Slash command

Policy enforcement (policy)

the gateway deciding, per request, whether this principal may call this tool on this server. The answer is allow, deny, or held until a named human approves it.

MCP gateways · See also MCP gateway, Audit log

Prefill (prefilling)

the phase that processes the whole prompt in one parallel pass, before any output token exists. Its cost grows with prompt length, which is why a large prompt is slow to start.

Inference and sampling · See also Decode, Time to first token

Progressive disclosure

keeping only a name and a one-line description in the context, and loading the body only when it is needed, so many skills cost little until used.

Skills, plugins, and MCP · See also Skill

Prompt (prompts)

the text sent to a model for it to continue, including everything the harness added that you never typed.

Start here · See also System prompt, Context window

Prompt caching (prompt cache, context caching)

a provider feature that stores the processed form of a request prefix for a short time and charges a reduced rate to reuse it on a later call.

Context management and compaction · See also Cache prefix

Q

Quantization (quantized, quantised, quantisation)

storing each weight in fewer bits, such as 4-bit instead of 16-bit, to cut memory and raise speed. The cost is accuracy, and the damage is uneven.

Local and cloud inference · See also Weights, Local inference

R

Rate limiting (rate limit, rate limits)

capping or queueing requests, both to stay inside a vendor's own limits and to stop one team consuming the whole organisation's quota.

LLM gateways and routers · See also Virtual key, Spend cap

Reasoning effort (thinking budget, thinking level)

the dial that sets how much working a reasoning model does, expressed either as a coarse level such as low or high, or as a target token budget.

Reasoning models · See also Reasoning model, Thinking tokens

Reasoning model (reasoning models, thinking model)

a model trained to produce a long private working before its answer, and to check and correct itself while doing so, rather than one prompted into showing its steps.

Reasoning models · See also Thinking tokens, Reasoning effort, Chain of thought

Recall

reading a memory store back into the context, usually at the start of a session and usually with no query to match it against.

Memory · See also Persistent memory, Semantic search

Reranking (reranker, rerank)

scoring retrieved candidates with a second, small model that reads the query and the chunk together, then keeping only the best few.

Embeddings, vector stores, and RAG · See also Top-k, Hybrid search

Retrieval-augmented generation (RAG)

retrieving relevant text from a corpus and putting it in the context window before asking the model to answer, so the answer rests on documents the model was never trained on.

Embeddings, vector stores, and RAG · See also Chunking, Knowledge base

S

Sampling (sampler, samplers)

the step outside the model that picks one token from the next-token distribution. It lives in the serving stack rather than in the weights, and it is where the visible randomness comes from.

Inference and sampling · See also Temperature, Top-p, Greedy decoding

Sandbox (sandboxing, sandboxed)

an enforced boundary around tool execution, covering which files a tool may read or write and which hosts it may reach. It holds whatever the model has been persuaded to try.

The agentic loop · See also Permission model

Semantic cache (semantic caching)

a cache that matches an incoming prompt against stored prompts by embedding distance rather than by exact bytes. It hits far more often than an exact-match cache and can answer a question that was not asked.

LLM gateways and routers · See also LLM gateway

finding text by distance between embeddings rather than by matching words, so a paraphrase with no shared vocabulary can still be found.

Embeddings, vector stores, and RAG · See also Hybrid search, Embedding

Server registry (registry, approved server list)

the reviewed list of MCP servers an organisation has approved, with an owner and a pinned version for each. It turns adding a server from a pasted config line into an act with a name on it.

MCP gateways · See also MCP gateway, Policy enforcement

Session memory

the conversation history, resent in full on every turn and discarded when the session ends.

Memory · See also Persistent memory, Message

Skill (skills, agent skill, agent skills)

a named set of instructions, packaged as markdown, that a harness loads into the context when it looks relevant. It adds no new capability.

Skills, plugins, and MCP · See also Progressive disclosure, Plugin

Slash command (slash commands)

an instruction the user invokes by name, rather than one the model chooses to load on its own.

Skills, plugins, and MCP · See also Skill, MCP prompt

Spend cap (spend caps, budget)

a budget enforced on the request path, which refuses calls once it is exhausted. A cap anywhere off the path can only report an overrun after the money is gone.

LLM gateways and routers · See also Virtual key, LLM gateway

stdio transport (stdio)

a server run as a local subprocess, exchanging newline-delimited JSON-RPC over its standard input and output, with credentials taken from its environment.

MCP in depth · See also Streamable HTTP transport

Stop condition (stop conditions)

a limit that ends an agentic run whether or not the job is finished, such as a turn cap, a token budget, a wall clock timeout, or a user interrupt.

The agentic loop · See also Agentic loop, Turn

Streamable HTTP transport (streamable HTTP)

a server reached over HTTP, where each message is a POST to one endpoint and the reply is a JSON object or a request-scoped stream.

MCP in depth · See also stdio transport

Streaming

sending each output token to the caller as it is decoded, rather than holding the whole response. Total generation time is unchanged; the wait is just spent reading.

Inference and sampling · See also Decode, Time to first token

Structured output (structured outputs, JSON mode)

forcing the model's reply to match a JSON Schema, with no tool and no execution involved. The same schema machinery pointed at the answer rather than at an action.

Tool calls · See also Tool definition

Subagent (subagents)

a nested agentic loop with its own context window, given one task and returning only its result, so a large search does not fill the main conversation.

The agentic loop · See also Agentic loop, Harness

System prompt (system prompts, system instruction)

the block of instructions an application or harness puts in front of the conversation, setting the model's role, its rules and the shape of its output. The reader usually never sees it.

What fills a context window · See also Context assembly, Message

T

Temperature

a sampler setting that flattens or sharpens the whole next-token distribution before a token is picked. Below 1 the leading candidate pulls further ahead; above 1 the also-rans get a real share.

Inference and sampling · See also Sampling, Top-p

Test-time compute (inference-time compute)

spending more computation at the moment of answering rather than at training time, as a way of buying accuracy. A smaller model thinking hard can beat a larger one answering immediately.

Reasoning models · See also Reasoning model, Reasoning effort

Thinking tokens (thinking token, reasoning tokens)

the tokens of a reasoning model's private working. They are billed at the output rate and occupy the context window whether or not the provider shows them to you.

Reasoning models · See also Output tokens, Reasoning effort

Time to first token (TTFT)

the delay before the first output token reaches the caller, driven mostly by prompt length and queueing rather than by how long the answer will be.

Inference and sampling · See also Prefill, Streaming

Token (tokens, tokenisation, tokenization)

the unit a model reads and writes, a chunk of a few characters drawn from a fixed vocabulary. Common English words are often one token; rare names, long numbers and punctuation-dense text cost several.

Tokens and the context window · See also Tokenizer, Context window, Input tokens, Output tokens

Tokenizer (tokenizers, tokeniser)

the software that splits text into tokens, using a vocabulary learned once from a corpus and frozen before training. Each model family has its own, so a token count is only meaningful against a named model.

Tokens and the context window · See also Token

Tokens per second (throughput, tok/s)

throughput. It means either what one request feels like as text appears, or what the machine produces across every request at once, and the two differ a lot.

Local and cloud inference · See also Continuous batching, Inference server

Tool (tools)

a capability the caller is willing to run on the model's behalf, offered to the model as a name, a description, and a schema.

Tool calls · See also Tool definition, Tool call

Tool call (tool calls, tool use)

a structured block the model emits naming a tool and its arguments. It executes nothing; the caller decides whether to run it.

Tool calls · See also Tool result, Function calling

Tool definition (tool definitions)

the name, description, and JSON Schema sent in the context so the model knows a tool exists. Every definition costs tokens on every call.

Tool calls · See also Tool, Structured output

Tool namespacing (namespacing)

prefixing each tool with the server it came from, so two servers exposing a tool of the same name can both be presented to one client. The prefix is text the model reads, so it changes behaviour as well as resolving the collision.

MCP gateways · See also MCP gateway, Server registry

Tool result (tool results)

a tool's output, appended to the conversation as a message and sent back to the model. It is text, and it costs context like any other text.

Tool calls · See also Tool call

Top-k (top k)

how many nearest chunks a search returns. A small k keeps the window clean; a large k refills it with noise.

Embeddings, vector stores, and RAG · See also Reranking

Top-p (nucleus sampling, top_p)

a sampler setting, also called nucleus sampling, that keeps only the smallest set of candidates whose probabilities sum to p and discards the rest before picking.

Inference and sampling · See also Sampling, Temperature

Trace (traces, tracing)

the stored record of one request through a gateway, holding the prompt, the routing decision, every attempt including the failed ones, the tokens counted, and the response.

LLM gateways and routers · See also LLM gateway, Audit log

Truncation (truncate)

dropping the oldest messages until a request fits, with no summary and no record of what was removed.

Context management and compaction · See also Compaction

Turn (turns)

one pass through the agentic loop. One model call, plus whatever the harness does with the answer before the next call.

The agentic loop · See also Agentic loop, Stop condition

V

Vector store (vector stores, vector database)

a database that holds embeddings beside the text they came from and answers nearest-neighbour queries against them, usually through an index rather than a full scan.

Embeddings, vector stores, and RAG · See also Embedding, Knowledge base

Virtual key (virtual keys)

a credential the gateway issues and controls, held by an application and never seen by the model vendor. It can be revoked, capped or attributed without touching the vendor's own key.

LLM gateways and routers · See also Spend cap, Rate limiting

W

Weights (model weights, parameters)

the file of numbers a training run produced, holding everything the model learned. Useless without code to run them, and the part that cost the money to make.

Open and closed models · See also Open-weights model, Quantization