Glossary

Every Algenta term in one place — the engine, the surfaces, decision planning, governed data, agentic runs, deployment, and licensing vocabulary, each defined once.

The canonical vocabulary for Algenta. Each term is defined once here; the rest of the docs use these words exactly. If a page and this glossary ever disagree, the glossary wins.

The engine

Term	Meaning
Engine	The Algenta core — a deterministic computational-intelligence runtime (math/operational libraries, simulation, decision planning, governed query, agent runtime). You deploy it; you don't adopt it as a finished app.
Runtime	The compiled compute capability integrated into an Algenta deployment.
Mojo engine	Algenta's compiled, high-performance numerical runtime.
Bundled runtime	The signed compiled compute component included with Algenta and supervised as part of the service. Users do not install or manage it separately.
Decision envelope	The structured result of a simulation or decision — expected value, confidence, percentiles, and a recommended action.
Determinism	The same inputs + seed always produce the same result. Algenta computation is reproducible by design.

Surfaces — how you reach the engine

Term	Meaning
HTTP API	The versioned REST surface under /v1. The default integration path; no SDK required. See API overview.
SDK (algenta-sdk)	The published Python HTTP client (decision_engine.AlgentaClient) and TypeScript/JavaScript client (AlgentaClient) for the API. See Python SDK and TypeScript SDK.
MCP server (algenta-mcp)	A Model Context Protocol bridge that exposes the engine as tools for Cursor, Claude Desktop, and other AI clients. See MCP server.
CLI	The de terminal client for the Algenta HTTP API. See CLI reference.

Decision planning

Term	Meaning
Simulation	A Monte Carlo run over a scenario that returns a decision envelope (POST /v1/simulate).
Distribution	The probabilistic shape of an input variable (for example normal, lognormal, triangular, beta, poisson).
Decision	A single recommended action plus its expected value and confidence.
Recommendation / Score / Compare	Decision primitives: rank actions, score one scenario, or compare alternatives.
Decision plan	A structured, multi-stage decision the engine produces and can later execute.
Decision memory	The persisted log of decisions and their later real-world outcomes, so models can learn from results.
What-if	Sensitivity analysis that varies inputs to see how the decision changes (POST /v1/simulate/what-if).

Governed data

Term	Meaning
Governed data	Data the engine queries under a typed, validated contract — schema-checked, with explicit field hints and operators — so results are deterministic and auditable.
Source	A registered, queryable data origin (schema-inferred and tracked in the source registry).
Dataset	An onboarded, engine-managed table you can summarize and query.
Connector	A configured connection to an external system (SQL, warehouse, REST, file) that backs a source or dataset.
Query / Resolve	A deterministic exact query (POST /v1/query) and the plan-resolution step that precedes it.
Capability plane	The unified router that turns an objective into the right provider/skill and executes it.
Capability / Skill	A discrete operation the capability plane can route to and execute (an LLM call, an optimizer, a data operation).

Agentic

Term	Meaning
Agent run	A persisted, observable agent execution with events, checkpoints, telemetry, approval gates, and lifecycle control.
Checkpoint	Saved agent state you can pause at, resume from, or replay.
Repository intelligence	The code-aware workflow: snapshot a repo, triage an issue, plan, simulate a patch, and apply a fix/PR.
Blast radius	The set of files and symbols transitively affected by a change — what depends on it. Drives graph queries, patch-apply gating, and adaptive CI scan scoping.
Blast-radius-scoped CI	Standard CI runs scoped by blast radius: pull requests whose diff cannot influence the test graph skip the pytest legs (guard, lint, and build still run); push, schedule, and dispatch always run exhaustive. See Use Algenta in CI/CD.
Async job	A long-running execution submitted to /v1/jobs, polled for status, and fetched on completion.
Trigger	A registered condition that fires an action (a webhook-driven automation).

Deploy & operate

Term	Meaning
Deployment mode	The privacy/egress profile: saas, vpc, self_hosted, air_gapped, or hybrid_provider. Set with ALGENTA_DEPLOYMENT_MODE.
Self-hosted	You run the full stack (API + worker + engine + Postgres + Redis + object storage) on your own infrastructure. See Self-hosting.
Air-gapped	A self-hosted profile with zero outbound network — licensing and execution are fully local.
Control plane	The hosted, engine-free surface (auth, accounts, teams, billing, licensing). A deployment runs full or control_plane via ALGENTA_SERVICE_ROLE.
Edition	A licensed product surface over the engine, selected with ALGENTA_PRODUCT.
Metering / Credits / Quota	Generic managed-runtime compatibility controls. Algenta editions do not use them for billing or execution caps; local activity records remain customer-owned. See Usage & execution policy.
Egress / Privacy report	Outbound-network policy and the operator proof (GET /v1/admin/privacy-report) that a deployment is cloud-free.

Security & licensing

Term	Meaning
API key	The bearer credential (Authorization: Bearer …) that authenticates requests. See Authentication.
Device binding	Binding a capped key to a specific device identity, enforced by fingerprint headers + a replay token.
License	A signed, offline-verifiable JWT (RS256) that grants product entitlements and capacity. Always v2. See Licensing.
Entitlement	What a license grants — products, features, and capacity caps (seats, nodes, workers).
BYOK	Bring Your Own Keys — your own provider credentials for LLM/AI capabilities, encrypted per tenant.
Execution ownership	Who runs a capability: algenta_managed (the engine) vs client_managed (your adapter). Fail-closed when unresolved.

Related

Core conceptsWhat Algenta is