The vocabularyof running agents.
This category invented a lot of words quickly, and plenty of them get used to mean different things. These are the definitions we use, written to be useful whether or not you ever buy anything.
// Plain definitions, written to stand on their own without a product pitch.
Protocol and platform
Terms that apply across both products and the wider agent tooling ecosystem.
Model Context Protocol (MCP)
Full entryThe open protocol that lets an AI agent discover and call external tools in a standard way.
Control plane
A coordination layer that carries state and control signals without executing the work itself.
MCP tools
Full entryThe named capabilities an MCP server exposes, discovered at connect time and called during a session without bespoke integration code.
MCP server
A service that exposes a set of tools an MCP-compatible agent can discover and call.
MCP memory server
Full entryA service exposing memory operations as MCP tools, so any compatible agent can write and retrieve context with no per-tool integration.
Context engineering
Full entryDeciding what an agent has in front of it for a task: what to retrieve, what to leave out, and where it came from.
AI driving AI
Full entryOne agent supervising or orchestrating another, answering its decisions through the same interfaces a person would use.
Interactive CLI vs Agent SDK
Full entryTwo ways to reach the same model: a supervised session under the local subscription, or a programmatic entry point billed as metered API usage.
Device flow
A sign-in pattern where a terminal authorises through a browser rather than a pasted token.
Seat
The per-member, per-product grant that gives someone access to a product.
Tenant isolation
Keeping one organization data unreachable from another, enforced in storage rather than application code.
Agent oversight
Terms for watching and steering a coding agent that is running on a machine you own.
Permission gates
Full entryThe pause where a coding agent asks before taking an action, and waits until somebody answers.
Plan mode
Full entryA working mode where an agent proposes what it intends to do and waits for approval before acting on any of it.
AI coding agent observability
Full entrySeeing what a running agent is doing, what it is blocked on, and what it did afterwards, at the level of sessions and decisions.
Async coding agents
Full entryAgents that keep working when nobody is watching, which makes their blocking moments the thing that decides whether the time gets used.
Session
One continuous run of a coding agent on a machine, with its own activity and decisions.
Message injection
Sending an instruction into a session that is already running.
Remote control for AI coding agents
Full entryFollowing and steering a session running on your own machine from a different device, including answering what it stops on.
Session context injection
Full entryPutting a correction or a missing constraint into a running session, so the agent adjusts without the run being restarted.
Agent fleet
Full entryEvery session an organization has running at once, treated as one thing to observe rather than as unrelated processes.
Interrupt
Stopping a session current activity immediately, without answering its pending question.
Daemon
The lightweight local process that reports session activity and maintains the live link.
Agent memory
Terms for the context an agent accumulates and retrieves across sessions.
AI agent memory
Full entryPersistent context an AI agent writes while working and retrieves in later sessions.
Semantic search
Full entryRetrieval that matches meaning rather than exact words, comparing vector embeddings so vocabulary stops mattering.
Memory scoping
Full entryThe rule deciding who a stored memory is visible to, from one machine up to a whole organization.
Vector embedding
A numeric representation of text that places similar meanings near each other.
Relevance scoring
Ranking that reflects what still matters as a corpus ages, refreshed on a schedule.
Lifecycle cleanup
Automatic expiry of stale memories so a corpus stays worth searching.
Hybrid search
Full entryCombining keyword and semantic matching, so exact identifiers and loose descriptions both retrieve well.
Importance scoring
Full entryWeighting memories by how much they still matter, so ranking reflects value rather than recency alone.
Persistent memory for AI agents
Full entryContext that outlives a single session, so an agent accumulates knowledge instead of starting from nothing each time.
Memory deduplication
Full entryRecognising that a new memory says what an existing one already said, and merging rather than stacking near-identical copies.
Governance
Terms for how an organization controls who gets in, what they can do, and how long data lives.
SSO and SAML
Full entryAuthenticating people through your own identity provider, so access policy is controlled centrally rather than per application.
SCIM provisioning
Full entryLetting the directory create, update, and deactivate accounts automatically, so offboarding stops depending on somebody remembering.
RBAC
Full entryRole-based access control: permissions attach to named roles, so what somebody can do follows from their role.
Data retention policy
Full entryHow long each kind of data is kept and what happens at the end of that period.
Audit log
An attributable record of who did what and when, kept so an action can be reconstructed afterwards.
Once the words make sense.
Start from a problem
The use-case pages are organised by the situation you are in rather than by product name.
Start from your team shape
The solutions pages cover teams, enterprise, agencies, and solo developers separately.
Start from your tools
The integrations directory covers which coding tools work with which product.
Definitions are free. So is deciding later.
When you are ready, an organization, a subscription, and one seat is the whole setup.