Glossary
Defined terms
Words the site uses with a specific meaning. Across the site, the first use of a term on a page is underlined and links here.
- Access profile
- What a member of an App may do: read, run, approve. Combined with a scope entitlement it becomes one grant.
- Where it is explained
- Agent Studio
- Where agent configuration is authored within the platform's guardrails. Skills brought from elsewhere are re-run through it.
- Agentic application (App)
- A full-stack business application in which people and agents work over the same pages, data, knowledge, and outputs, under one identity, one authorization model, and one audit record. On Pyrana an App is configuration on a platform that is built once.
- Where it is explained
- App document
- The one document that declares an application: its surfaces, targets, agents, workflows, tool sources, resources, access, and knowledge. Compiled into an immutable, content-digested version; activated by moving one pointer.
- Where it is explained
- Approval lane
- A named group or relation (the requesting user, a finance group, a workflow, a tool relation) that receives an independent obligation to approve. Reviewer eligibility is checked at decision time.
- Classification
- How carefully a Context Unit may change: absolute (never auto-changes), regulated (needs a human), contextual, tribal (conflict-checked), derived (updates itself).
- Where it is explained
- Context engine (cortIQ)
- The layer that holds company knowledge as typed, classified, evidence-backed Context Units in a knowledge graph, retrieves them for agents, and accounts for what each run retrieved, injected, and cited. cortIQ is Pyrana's context engine, formerly the Context Engine.
- Where it is explained
- Context search
- The mid-run call an agent makes to the context engine: it pauses, retrieves, and resumes. The same path serves the copilot.
- Where it is explained
- Context Unit (CxU)
- The unit of knowledge in cortIQ: one atomic claim, its verbatim supporting passage with source and location, a knowledge type, a classification, an evidence grade, a quality score, and an identity that is a content hash.
- Where it is explained
- Copilot
- The App's assistant, auxiliary to its pages: it answers questions about what is on screen with the same authorization as the person, and it is the pathway to personal pages. Not the primary interface.
- Door
- How a call arrives: the App's own pages, the copilot, the API, MCP from another agent, a schedule, or an external event. Every door leads to the same gate.
- Durable execution
- Work that survives waits, restarts, and deploys. Workflow lanes run on Temporal workers and resume from a checkpoint after any interruption.
- Where it is explained
- Engine-side agents
- Agents inside the context engine: a query agent that plans the graph walk and repairs its own queries, and a learning agent that scores what was retrieved and used and flags units that never help.
- Execution record
- The one record every call leaves: caller, App version, target, resource versions, authorization outcome, approvals, effects with receipts, and the context cited. Written before dispatch.
- Where it is explained
- Extension
- A client-owned durable backend service deployed in the same cluster, built with the platform library that handles authentication and authorization, reachable only through the platform. Used for custom business logic and storage.
- Where it is explained
- Frozen invocation
- The immutable proposal a reviewer approves: the operation, its exact arguments, scope, and evidence, under a hash. The waiting agent cannot alter it; it is revalidated before it dispatches once.
- Knowledge type
- What kind of claim a Context Unit is: axiomatic, empirical, prescribed, or procedural.
- Where it is explained
- Lane
- An execution mode in the Pyrana Harness: direct (deterministic, no model), agent, workflow (durable), batch, or event.
- Where it is explained
- MCP
- Model Context Protocol. Every App exposes an MCP server, so any agent (Claude, ChatGPT, a client's own) can work with the App as the user, under the user's own authentication and authorization.
- Page context
- What the platform hands the agent about the page a person is on: which page, which entity, which period, which resource versions.
- Personal page
- A page a person asks the copilot for. The copilot authors it from the resources and read capabilities the person may use; the backend compiles it with no network, imports, storage, or authority; the browser renders it with every read still authorized. Read-only today.
- Where it is explained
- Push, activate, deploy
- Push compiles the App document into an immutable version and registers the App identity. Activate moves the active pointer, is audited, and requires an App owner; rollback is an activation of a prior version. Deploy ships a frontend image and registers and activates nothing.
- Where it is explained
- Pyrana Harness
- The runtime and orchestration layer: native tool calling, planning, subagents, checkpoints and resume; durable workflows that fan out, wait, retry, and finish.
- Where it is explained
- Resource
- A versioned, immutable, attributed output of an App: v1 published by a run, v3 revised by the copilot on behalf of a person. Publishing against a stale version is refused. Pages read a resource at an exact version.
- Scope entitlement
- The slice of data a caller's profile applies to: an entity, a project, a review queue. What a caller may not see is absent from results, not denied.
- Standing set
- A set of Context Units preloaded before an agent's first turn, for example an entity's policies.
- Surface
- A declared page where a person works. The platform knows what the person is looking at and hands it to the agent as exact page context.
- Target
- An operation the App can perform, declared once and reachable from many doors through one gate.
- The gate (admission)
- The eight-step check every call passes before anything runs: identity and idempotency key; active App version; target exists and caller allowed; references resolved to exact versions; one authorization check; execution row and audit record written in one transaction; dispatch; outcome recorded. Fails closed.
- Where it is explained
- Tool approval policy
- A rule over an operation's typed arguments (thresholds, membership, existence, compound rules) that decides whether a person must approve before it runs.
- Where it is explained
- Tool source
- A connection an App binds to: REST, ODBC, MCP and MCP gateways, Databricks, Snowflake, Google Drive, OneDrive. Tools are specific to the job; the agent passes the entity and the period, the tool holds the query.