diff --git a/assets/js/usecase-views.js b/assets/js/usecase-views.js index 467fb0a..22db0df 100644 --- a/assets/js/usecase-views.js +++ b/assets/js/usecase-views.js @@ -1,6 +1,6 @@ /* Public page-load counters; previews never increment production totals. */ (() => { - const paths = ["/usecases/1/", "/usecases/2/", "/usecases/3/", "/usecases/4/", "/usecases/5/", "/usecases/6/", "/agentic-sidecar/usecases/prevent-unauthorized-refunds/", "/agentic-sidecar/usecases/governed-customer-support-agent/", "/agentic-sidecar/usecases/govern-langgraph-agent-workflows/", "/control-tower/usecases/fleet-health-snapshot/", "/control-tower/usecases/lens-failure-to-incident/", "/control-tower/usecases/ecosystem-evidence-posture/", "/control-tower/usecases/fleet-change-with-approval/"]; + const paths = ["/usecases/1/", "/usecases/2/", "/usecases/3/", "/usecases/4/", "/usecases/5/", "/usecases/6/", "/agentic-sidecar/usecases/prevent-unauthorized-refunds/", "/agentic-sidecar/usecases/governed-customer-support-agent/", "/agentic-sidecar/usecases/govern-langgraph-agent-workflows/", "/control-tower/usecases/fleet-health-snapshot/", "/control-tower/usecases/lens-failure-to-incident/", "/control-tower/usecases/ecosystem-evidence-posture/", "/control-tower/usecases/fleet-change-with-approval/", "/agentic-evals/usecases/block-an-unsafe-release/", "/agentic-evals/usecases/right-reply-wrong-trajectory/", "/agentic-evals/usecases/gate-on-judge-reliability/"]; const normalize = path => path.replace(/index\.html$/, '').replace(/\/?$/, '/'); const path = normalize(location.pathname); if (['deepagentlabs.io', 'www.deepagentlabs.io'].includes(location.hostname) && paths.includes(path)) { diff --git a/scripts/build_docs.py b/scripts/build_docs.py index 45e9dd1..9bcdd25 100644 --- a/scripts/build_docs.py +++ b/scripts/build_docs.py @@ -311,7 +311,9 @@ def source_link(match, page=page): }, "extra_css": [css] + figure_css, "extra_javascript": ( - ["legacy-links.js"] if project == "agentic-evals" else [] + ["legacy-links.js"] + if project == "agentic-evals" and section == "docs" + else [] ) + figure_js, "plugins": ["search"], diff --git a/scripts/update_usecase_views.py b/scripts/update_usecase_views.py index c494465..f0a6da0 100644 --- a/scripts/update_usecase_views.py +++ b/scripts/update_usecase_views.py @@ -17,6 +17,11 @@ "fleet-health-snapshot", "lens-failure-to-incident", "ecosystem-evidence-posture", "fleet-change-with-approval", ) +] + [ + f"/agentic-evals/usecases/{slug}/" for slug in ( + "block-an-unsafe-release", "right-reply-wrong-trajectory", + "gate-on-judge-reliability", + ) ] diff --git a/usecases/index.html b/usecases/index.html index 51b213e..7ccd63c 100644 --- a/usecases/index.html +++ b/usecases/index.html @@ -1,7 +1,7 @@ -AgenticOps use cases · DeepAgentLabs +AgenticOps use cases · DeepAgentLabs -
+
DeepAgentLabs @@ -96,4 +96,4 @@ Remote MCP GitHub -

DEEPAGENTLABS / USE CASES

From an agent run
to operational confidence.

See how the ecosystem comes together. Hands-on stories, one continuous loop: observe, test, evaluate, improve.

Interactive concept demos Fictional data · Runs in your browser · No setup needed
01 INTERACTIVE DEMO ↗
DeveloperHow are my agents doing?↓ MCP · Lens · Evals

CONNECT & OPERATE

Operate your agents through MCP

One conversation. Your entire AgenticOps ecosystem. Follow a support incident from the first question to a consolidated operations report.

Ask. Investigate. Understand. →— views02 INTERACTIVE DEMO ↗
Production run↻   Replay + inject faultTimeout → fallback → recovery

REPLAY & IMPROVE

Replay, stress-test, optimize

Turn a captured production run into a repeatable experiment. Introduce a failure, inspect the evidence, and validate a better recovery strategy.

Break it here. Improve it safely. →— views03 INTERACTIVE DEMO ↗
Baseline   →   Candidate78 → 94Quality improves. Safety still gates.

EVALUATE & COMPARE

Evaluate today. Validate your next release.

Score a recorded execution, then replay its scenario against a candidate release. See improvements and regressions before making a release decision.

Make the release decision visible. →— views04 INTERACTIVE DASHBOARD ↗
Orchestrator → subagents → tools▰ ▰▰   2,840 tokensDistributed tracing · Cost · Diagnostics

OBSERVE & DIAGNOSE

See inside every agent run.

Explore an AgenticLens observability console. Inspect nested traces, token usage by agent, model costs, retries and evidence-backed findings.

Follow the trace. Find the bottleneck. →— views05 ANIMATED SIMULATION ↗
Agent → Sidecar → external callsCheck. Review. Approve. Execute.Guardrails · Intent · Critique · Approval

SUPERVISE & EXECUTE

A checkpoint before every call.

Watch Sidecar supervise an agent’s tools, MCP and vector database calls. Automated checks precede every call; critical actions also require scripted human approval.

Watch the supervision loop. →— views06 INTERACTIVE FLEET CONSOLE ↗
Development · Staging · ProductionInspect → Review → ControlFleet evidence · Approvals · Emergency brake

OPERATE & GOVERN

An AI-native operations command center.

Investigate a failure with MCP + Copilot, approve a remediation runbook and verify recovery. Or stop an agent, block new calls and revoke access.

Take the operator’s seat. →— views

MORE USE CASES / AGENTIC-SIDECAR

Explore more Sidecar examples.

AGENTIC-SIDECARPrevent unauthorized refundsOpen use case — viewsAGENTIC-SIDECARGoverned customer support agentOpen use case — viewsAGENTIC-SIDECARGovern LangGraph agent workflowsOpen use case — views

MORE USE CASES / AGENTICOPS-CONTROL-TOWER

Explore more Control Tower examples.

AGENTICOPS-CONTROL-TOWERFind unhealthy agents and capability gapsOpen use case — viewsAGENTICOPS-CONTROL-TOWERTurn a failed AgenticLens run into an incidentOpen use case — viewsAGENTICOPS-CONTROL-TOWEROne posture view for every producerOpen use case — viewsAGENTICOPS-CONTROL-TOWERChange every staging agent with a second approverOpen use case — views

Page views since counters were added · Counts update in about five minutes · Powered by Hits

ONE STORY, MANY PERSPECTIVES

Meet Northstar Support.

A fictional support team uses a triage agent, an order specialist and a resolution agent to handle a delayed order. Explore how the team investigates a slow run, tests recovery and checks its next release.

Capture → Standardize → Replay → Stress → Evaluate → Improve

These stories illustrate a possible ecosystem workflow. Tool names, artifacts, scores and outcomes are illustrative; they do not demonstrate live integrations or guarantee current product support.

+

DEEPAGENTLABS / USE CASES

From an agent run
to operational confidence.

See how the ecosystem comes together. Hands-on stories, one continuous loop: observe, test, evaluate, improve.

Interactive concept demos Fictional data · Runs in your browser · No setup needed
01 INTERACTIVE DEMO ↗
DeveloperHow are my agents doing?↓ MCP · Lens · Evals

CONNECT & OPERATE

Operate your agents through MCP

One conversation. Your entire AgenticOps ecosystem. Follow a support incident from the first question to a consolidated operations report.

Ask. Investigate. Understand. →— views02 INTERACTIVE DEMO ↗
Production run↻   Replay + inject faultTimeout → fallback → recovery

REPLAY & IMPROVE

Replay, stress-test, optimize

Turn a captured production run into a repeatable experiment. Introduce a failure, inspect the evidence, and validate a better recovery strategy.

Break it here. Improve it safely. →— views03 INTERACTIVE DEMO ↗
Baseline   →   Candidate78 → 94Quality improves. Safety still gates.

EVALUATE & COMPARE

Evaluate today. Validate your next release.

Score a recorded execution, then replay its scenario against a candidate release. See improvements and regressions before making a release decision.

Make the release decision visible. →— views04 INTERACTIVE DASHBOARD ↗
Orchestrator → subagents → tools▰ ▰▰   2,840 tokensDistributed tracing · Cost · Diagnostics

OBSERVE & DIAGNOSE

See inside every agent run.

Explore an AgenticLens observability console. Inspect nested traces, token usage by agent, model costs, retries and evidence-backed findings.

Follow the trace. Find the bottleneck. →— views05 ANIMATED SIMULATION ↗
Agent → Sidecar → external callsCheck. Review. Approve. Execute.Guardrails · Intent · Critique · Approval

SUPERVISE & EXECUTE

A checkpoint before every call.

Watch Sidecar supervise an agent’s tools, MCP and vector database calls. Automated checks precede every call; critical actions also require scripted human approval.

Watch the supervision loop. →— views06 INTERACTIVE FLEET CONSOLE ↗
Development · Staging · ProductionInspect → Review → ControlFleet evidence · Approvals · Emergency brake

OPERATE & GOVERN

An AI-native operations command center.

Investigate a failure with MCP + Copilot, approve a remediation runbook and verify recovery. Or stop an agent, block new calls and revoke access.

Take the operator’s seat. →— views

MORE USE CASES / AGENTIC-SIDECAR

Explore more Sidecar examples.

AGENTIC-SIDECARPrevent unauthorized refundsOpen use case — viewsAGENTIC-SIDECARGoverned customer support agentOpen use case — viewsAGENTIC-SIDECARGovern LangGraph agent workflowsOpen use case — views

MORE USE CASES / AGENTICOPS-CONTROL-TOWER

Explore more Control Tower examples.

AGENTICOPS-CONTROL-TOWERFind unhealthy agents and capability gapsOpen use case — viewsAGENTICOPS-CONTROL-TOWERTurn a failed AgenticLens run into an incidentOpen use case — viewsAGENTICOPS-CONTROL-TOWEROne posture view for every producerOpen use case — viewsAGENTICOPS-CONTROL-TOWERChange every staging agent with a second approverOpen use case — views

MORE USE CASES / AGENTIC-EVALS

Explore more Evals examples.

AGENTIC-EVALSBlock a release when one safety case failsOpen use case — viewsAGENTIC-EVALSCatch a right reply reached the wrong wayOpen use case — viewsAGENTIC-EVALSDon't let a flaky LLM judge approve a releaseOpen use case — views

Page views since counters were added · Counts update in about five minutes · Powered by Hits

ONE STORY, MANY PERSPECTIVES

Meet Northstar Support.

A fictional support team uses a triage agent, an order specialist and a resolution agent to handle a delayed order. Explore how the team investigates a slow run, tests recovery and checks its next release.

Capture → Standardize → Replay → Stress → Evaluate → Improve

These stories illustrate a possible ecosystem workflow. Tool names, artifacts, scores and outcomes are illustrative; they do not demonstrate live integrations or guarantee current product support.