Glossary
This page defines the terms used across the documentation. Each term has its own link, so other pages can point to it.
Account
Section titled “Account”Your Alvia account, shown as your organization in the dashboard. It has one plan, and its rate limits and allowances cover all its projects.
Adaptive Relevance
Section titled “Adaptive Relevance”How Alvia ranks memories: a boost set by each memory’s category that fades with its age, plus favorites, recent use and sentiment. Similarity to the query still leads. See Temporal decay and Age and temporal decay.
Allowance
Section titled “Allowance”How many reads and writes a plan includes each month. For stored memories, see Stored-memory limit and Limits and allowances.
API key
Section titled “API key”A secret that starts with sk-mem- and gives access to one project. See
Authentication.
Category
Section titled “Category”The kind of fact a memory holds: identity, preference, goal, commitment, decision,
constraint, relationship, work, knowledge, skill, resource, event, metric or
issue. New categories can be added.
A part of an imported conversation. Alvia splits long conversations into chunks, and each chunk uses one write.
Classification
Section titled “Classification”How formation classifies a new fact against what Alvia holds. Each fact gets one outcome: Supersession, Refinement, Duplicate, Contradiction, Related or Novel. Each version records its outcome with the classifier’s explanation, which inspect returns and the dashboard shows in the memory’s history. See Memory formation.
Context
Section titled “Context”A short answer, one to three sentences, that Alvia resolves from the memory that best matches a
query and the memories related to it. Returned by POST /v1/memories/context and the context
tool.
Contradiction
Section titled “Contradiction”Two memories that conflict, where neither replaces the other. Alvia keeps both and links them.
Demo Project
Section titled “Demo Project”A project every new account starts with: about 25 sample memories about Tidepool, a fictional habit-tracking app, including replaced facts and a contradiction. It does not count toward the project limit, and its sample memories do not count toward the stored-memory limit.
Duplicate
Section titled “Duplicate”A fact Alvia already holds. Formation stores nothing new for it.
Entity
Section titled “Entity”A person, product, project, tool, date or other thing a memory names. Each memory lists its entities with a weight from 0 to 1.
Favorite
Section titled “Favorite”A memory you marked with is_favorited: true. It ranks higher in reads. See
Manage memories.
Forget
Section titled “Forget”Remove a memory from every read, permanently. It stops counting against the stored-memory limit. Its text, versions, relationships and feedback stay stored until the project is deleted. See Manage memories.
Formation
Section titled “Formation”The background process that turns a write into memories: it takes out facts, compares them with existing memories, classifies each one and stores the result. See Memory formation.
Full Lineage
Section titled “Full Lineage”A memory’s versions, newest first, each with the outcome that produced it and why, and its relationships to other active memories. Inspect returns it, and the dashboard shows it in the memory’s History and Relationships.
The write option that reads content as a passage and forms a memory from each fact in it. It is
on by default.
Inspect
Section titled “Inspect”A read that returns one memory with its versions, how Alvia classified each one and why, and its relationships to other active memories. See Inspect a memory.
The Model Context Protocol, a standard way for AI agents to call tools. Alvia runs an MCP server. See Connect a coding agent with MCP.
Memory
Section titled “Memory”One durable fact that Alvia holds, with an ID that starts with mem_.
Memory policy
Section titled “Memory policy”Project-level rules that exclude new memories by keyword or meaning. They are separate from read filters, access control and deletion. See Memory policies.
A new fact that relates to nothing Alvia holds. Formation stores it as a new memory.
Occurrence time
Section titled “Occurrence time”The time a memory’s fact was stated or observed, in the occurred_at field. See
Point-in-Time Reasoning and change over time.
Passage
Section titled “Passage”Text with one or more facts in it. With infer on, Alvia forms a memory from each fact in a
passage.
Playground
Section titled “Playground”A page for each project in the dashboard that runs List, Search, Recall, Context and Create calls against the project and shows each one as code. Its calls count against the read and write allowances. See Limits and allowances.
Point-in-Time Reasoning for AI™
Section titled “Point-in-Time Reasoning for AI™”Alvia’s product category, not the category field (see Category). At write time Alvia
resolves how each new fact relates to what it holds (it replaces, adds detail to, repeats,
contradicts or relates to it, or is new), keeps the lineage and answers from what applies now, using
when each fact was stated. See Point-in-Time Reasoning and change over
time.
Project
Section titled “Project”A separate set of memories, with its own API keys. See Projects and scope.
Ranked score
Section titled “Ranked score”A number from 0 to 1 that shows how relevant a memory is to a query. Compare it only within one response.
A request that returns memories or context: context, recall, search, get, inspect or list. Context uses five reads from your monthly allowance, inspect uses three and every other read uses one.
Recall
Section titled “Recall”The read that returns the single best memory for a query. See Read memories.
Refinement
Section titled “Refinement”A new fact that adds detail to a memory. Alvia merges it into the memory’s next version.
Related
Section titled “Related”A new fact that relates to a memory without changing it. Alvia stores it as a new memory and links the two.
Five optional labels, such as user_id and agent_id, that say who or what a memory is about.
See Projects and scope.
Search
Section titled “Search”The read that returns a ranked list of memories for a query. See Read memories.
Sentiment
Section titled “Sentiment”Feedback on a memory: positive (useful), negative (wrong, outdated or irrelevant) or neutral.
It changes how the memory ranks.
Stored-memory limit
Section titled “Stored-memory limit”How many active memories a plan can hold. It does not reset each month. See Limits and allowances.
Supersession
Section titled “Supersession”A new fact that replaces an earlier one. It becomes the memory’s next version, and reads return it instead of the earlier fact.
Temporal decay
Section titled “Temporal decay”A small ranking boost, set by a memory’s category, that fades with its age: quickly for events, slowly for preferences. An old memory that matches a query well still ranks. Decay is not deletion. See Point-in-Time Reasoning and change over time.
Temporal Graph
Section titled “Temporal Graph”A project’s memories, each with its dated versions, and the relationships between them: related memories and contradictions. A fact that replaces a memory or adds detail to it becomes its next version. The dashboard’s Graph shows it. See Version.
Token bucket
Section titled “Token bucket”How Alvia limits request rates. Each request takes a token, and the bucket refills continuously at the plan’s rate. See Rate limits.
Version
Section titled “Version”One state in a memory’s forward evolution. A memory gets a new version when a fact supersedes or refines it. Reads return the current version, and every version stays in the memory’s history, available through the API and in the dashboard. See Point-in-Time Reasoning and change over time.
A request that sends information for formation: store or import. A store request, or each chunk of an imported conversation, uses one write from your monthly allowance, and so does each fact formed from it.