SerenQuantDocs
Sign up
Sign up
© 2026 SerenQuant. All rights reserved.
DocsPrivacyTermsStatusSupport
Docs
Documentation
Start here
Overview
Guide library
GuidesGetting Started: APIGetting Started: MCPAuth + Workspace ScopesAccount and Workspace ManagementMarkets WorkspaceMarkets Symbol WorkspaceMarket Guides and Beta ScopeFutures ReadinessResearch WorkflowResearch Workbench, Pipeline, and ExperimentsResearch Specialist MethodsResearch Context SurfacesBacktests Run LifecycleBacktests Run Detail and CompareStrategy Sweeps and OptimizerExecution WorkflowProviders and BenchmarkingSettings and Runtime ConfigAgent WorkbenchNews WorkflowDatasets WorkflowGlobal Language SelectionMFA Getting StartedMFA Recovery + Device LossMFA Trusted DevicesMFA API ReferenceLLM Market Analysis LoopStrategy Generation + Backtest LoopPromotion + Risk Guardrails
Reference
API ReferenceMCP Reference
Lifecycle
Changelog
  1. Docs/
  2. Guides/
  3. Agent Workbench
Public docs

Guide: Agent Workbench

Understand the signed-in agent surface across copilot threads, scheduled automations, delivery routing, and workflow-console operations.

Signed-in workflow

What this guide covers

Use this guide when you need to understand how the Agent Workbench combines persistent copilot threads, scheduled automations, delivery, and workflow orchestration into one workspace-bound surface.

  • Treat the agent page as an operating surface for ongoing research execution, not just a chat entry point.
  • Copilot, automations, delivery, timeline, and workflow-console sections are designed to work together so evidence and handoffs stay visible.
  • Keep MCP, policy, and workspace context in mind because agent workflows inherit the same guardrails as the rest of the signed-in product.

Routes and surfaces

The current Agent Workbench page is organized as a sequence of operational surfaces rather than separate routes.

/agent · copilot workspace

Conversational research copilot

Use persistent workspace-bound threads with provenance, citations, and follow-up history.

/agent · scheduled automations

Scheduled automations

Create recurring briefs, alerts, drift summaries, and promotion-gate checks with profile-aware delivery.

/agent · delivery routing & inbox

Delivery routing and inbox

Configure where outputs land, then triage automation and copilot results from a shared inbox surface.

/agent · activity timeline

Activity timeline

Audit copilot threads, automation runs, tool calls, delivery events, and policy actions in one explainability rail.

/agent · workflow console

Workflow console

Run structured MCP-native workflows and review orchestration history plus guardrail detail.

Recommended agent loop

Use this sequence to keep agent workflows persistent, reviewable, and policy-aware.

01

Start with a copilot thread

Use a workspace-bound thread to frame the task, preserve provenance, and keep follow-up context stable.

02

Promote stable tasks into automations

Once a thread turns into a recurring job, formalize it as an automation instead of re-running it ad hoc.

03

Route outputs intentionally

Set delivery and inbox behavior so the right workspace or user sees the output without losing traceability.

04

Audit with timeline and workflow console

Use the timeline and workflow console to confirm tool use, policy events, and orchestration history before trusting the result.

Copilot, automation, and delivery controls

The first half of the workbench is where research intent becomes scheduled, routed work.

thread memory

Persistent copilot threads

Start from workspace-bound thread history so follow-up prompts inherit prior answers, provenance, and citations instead of restarting the context each time.

scheduled work

Automation definitions and profile-aware runs

Use the automation center to turn stable prompts into recurring jobs with workspace-aware cadence and profile assumptions.

delivery routing

Workspace and personal delivery channels

Configure where agent outputs land before you scale a workflow. Delivery routing is part of the workflow contract, not an afterthought.

result triage

Inbox review and handoff

The delivery inbox is where generated output gets triaged, acknowledged, and handed into research or execution workflows.

Timeline and workflow-console controls

The second half of the workbench is where you verify what the agent actually did.

activity log

Copilot, automation, and delivery timeline

Use the timeline to see threads, automation runs, tool calls, and delivery events in one place before trusting the result.

orchestration

Workflow console and guardrails

The workflow console is the structured execution surface for MCP-native workflows. It is where run history and guardrail detail stay visible.

policy trace

Policy and tool events

Check policy and tool events when output quality changes unexpectedly. They explain whether the issue came from permissions, routing, or execution.

handoff evidence

End-to-end handoff trace

Keep the trace from prompt to delivery intact so downstream teams can see how an agent result was produced.

Example agent operations

Use these scenarios when a thread or automation needs a concrete operator playbook instead of a generic walkthrough.

scheduled brief

Turn a successful copilot thread into a morning brief

Start from the thread that already has the right workspace context, convert it into an automation with a named cadence, and route the result into a delivery channel the desk actually reviews each morning.

Expected outcome

The recurring brief inherits the original thread context and arrives in a predictable place without re-prompting from scratch.

guardrail escalation

Investigate a sudden guardrail or tool failure

Use the activity timeline and workflow console together to locate the failed run, inspect the policy and tool events around it, and confirm whether the problem came from permissions, MCP availability, or prompt drift.

Expected outcome

You end with a specific failure source and a traceable record before rerunning or escalating the automation.

handoff trace

Audit a delivered output before it enters research or execution

Open the inbox item, match it back to the originating thread or automation, then confirm the delivery channel, timeline entries, and workflow-console history all point to the same run.

Expected outcome

The downstream consumer gets an output with intact provenance instead of an orphaned summary.

Guardrails and review points

  • Do not treat a useful copilot result as durable until the workspace thread and provenance trail are both intact.
  • Convert recurring work into automations only after the delivery route and profile assumptions are explicit.
  • Use the activity timeline and workflow console to inspect policy or tool events before escalating an agent output into broader workflows.
  • Keep MCP and settings posture in sync with agent behavior so the workbench does not drift beyond current workspace controls.

Next steps

Open Agent Workbench

Jump into the signed-in agent surface and review copilot, automations, delivery, and workflow history.

MCP Reference

Cross-check the MCP tools, resources, and bridge contracts that agent workflows rely on.

Settings guide

Review the runtime and agent-profile controls that shape automation and copilot behavior.

Last updated

Mar 24, 2026

Feedback

Report unclear guidance, stale contracts, missing coverage, or broken docs UI on this page.

Open feedback issue
Previous
Settings and Runtime Config
Next
News Workflow
On this page

Jump to the section you need without losing your place.

  • What this guide covers
  • Routes and surfaces
  • Recommended agent loop
  • Copilot, automation, and delivery controls
  • Timeline and workflow-console controls
  • Example agent operations
  • Guardrails and review points
  • Next steps
Last updated

Mar 24, 2026

Feedback

Report unclear guidance, stale contracts, missing coverage, or broken docs UI on this page.

Open feedback issue