Core concepts
The Veyra platform is built around five primitives. Understand them and you understand the whole system.
Workspace
The top-level container. Holds members, billing, secrets, and policy. All runs are scoped to a workspace.
Run
A single autonomous task triggered by a user, the API, or a webhook. A run has a goal, a plan, an execution trace, and an outcome.
Agent
A specialised reasoning loop with access to a fixed toolset. Veyra ships with four primary agents — Planner, Coder, Reviewer, Operator — and 47 sub-agents that the primaries delegate to.
Provider
Underlying LLM endpoint. Veyra selects providers per task using cost, latency, capability, and your routing policy.
Trace
An immutable, structured log of every tool call, model invocation, and file diff produced by a run. Traces power debugging, audit and replay.
How they fit together
| Layer | Owns | Lives in |
|---|---|---|
| Workspace | Members, billing, policy | Settings |
| Run | Goal + outcome | Run history |
| Agent | Planning + tool choice | Agents page |
| Provider | Model invocation | Providers page |
| Trace | Audit + replay | Logs |
Mental model
- You give a workspace a goal — that creates a run.
- The Planner agent breaks the goal into steps and delegates to sub-agents.
- Each step routes to the optimal provider.
- Everything is recorded in a trace you can replay or audit.