Capabilities

Studio discovers its surfaces from registered targets and configured stores.

Surface	Package behavior
Playground	Runs real agents, streams JSONL events, stores transcripts, supports cancel, approvals, questions, attachments, and model selection
Agents	Reports identity, model, tools, context, memory, middleware, hooks, limits, and runtime metadata
Tools	Shows schemas and metadata and can invoke registered tools directly
MCP	Shows server/tool origin and can invoke MCP-backed tools
Pipelines	Exposes graphs, typed input metadata, runs, logs, stored history, and replay
Sessions	Creates, lists, updates, loads, and deletes Studio chat sessions
Traces	Records local run/generation/tool trees with status, timing, usage, and request context
Knowledge	Inspects static context, dynamic context, dynamic tools, and retrieval evidence
Memory	Uses an agent's read-only MemoryInspector when available, otherwise Studio session storage
Sandboxes	Discovers sessions attached to sandbox tools and exposes read-only files, ports, processes, and bounded logs
Status	Reports enabled capabilities, stores, registered targets, and available record counts
Evaluations	Registers typed suites and exposes evaluation configuration/run HTTP routes

Capability limits

Studio can only show what the runtime exposes. Knowledge inspection is not a document editor. Memory adapters without an inspector cannot expose their durable conversations. Sandbox inspection is read-only, although the registered agent tools may still mutate the sandbox. Runtime status reports configured capabilities, not dependency health checks.

Evaluation runtime routes exist for registered suites; the current primary browser navigation is centered on Playground, Pipelines, Sessions, Traces, and inspection surfaces rather than a dedicated Evaluations page.

Studio traces are local development evidence. Use Lens for retained team observability and evaluation workflows across deployments.

See runtime boundary and the API reference.