Skip to content

README.md

Latest commit

 

History

History
158 lines (115 loc) · 6.09 KB

README.md

File metadata and controls

158 lines (115 loc) · 6.09 KB

Examples

Top level:

  • resources/ — manifests grouped by resource kind (agents, tasks, tools, …). See resources/README.md.
  • blueprints/ — minimal pipeline / hierarchical / swarm topology templates.
  • use-cases/ — copy-paste scenario bundles.

Layout

  • resources/agents/
  • resources/agent-systems/
  • resources/model-endpoints/
  • resources/tools/
  • resources/memories/
  • resources/secrets/
  • resources/agent-policies/
  • resources/agent-roles/
  • resources/tool-permissions/
  • resources/tasks/
  • resources/task-schedules/
  • resources/task-webhooks/
  • resources/workers/
  • resources/mcp-servers/
  • blueprints/
  • use-cases/

Real-world scenarios

Self-contained use case directories (full YAML templates + README per scenario) live under:

  • examples/use-cases/README.md — index
  • examples/use-cases/weekly-intelligence-brief/, cross-functional-pmo/, roadmap-synthesis-swarm/, event-driven-webhook/

Quick Start (Base Flow)

go run ./cmd/orlojctl apply -f examples/resources/memories/research_memory.yaml
go run ./cmd/orlojctl apply -f examples/resources/model-endpoints/openai_default.yaml
go run ./cmd/orlojctl apply -f examples/resources/tools/web_search_tool.yaml
go run ./cmd/orlojctl apply -f examples/resources/tools/vector_db_tool.yaml
go run ./cmd/orlojctl apply -f examples/resources/secrets/search_api_key.yaml
go run ./cmd/orlojctl apply -f examples/resources/secrets/openai_api_key.yaml
go run ./cmd/orlojctl apply -f examples/resources/agents/planner_agent.yaml
go run ./cmd/orlojctl apply -f examples/resources/agents/research_agent_model_ref.yaml
go run ./cmd/orlojctl apply -f examples/resources/agents/writer_agent.yaml
go run ./cmd/orlojctl apply -f examples/resources/agent-systems/report_system.yaml
go run ./cmd/orlojctl apply -f examples/resources/agent-policies/cost_policy.yaml
go run ./cmd/orlojctl apply -f examples/resources/tasks/weekly_report_template_task.yaml
go run ./cmd/orlojctl apply -f examples/resources/task-schedules/weekly_report_schedule.yaml
go run ./cmd/orlojctl apply -f examples/resources/secrets/webhook_shared_secret.yaml
go run ./cmd/orlojctl apply -f examples/resources/task-webhooks/generic_webhook.yaml

Starter Blueprints

For reusable architecture templates (pipeline, hierarchical, swarm+loop), see:

  • examples/blueprints/README.md

For personal runtime verification scenarios (including retry/deadletter and governance deny paths), see:

  • testing/scenarios/README.md

For live-provider runtime scenarios (real model credentials required), see:

  • testing/scenarios-real/README.md

Model routing is configured per agent via spec.model_ref. Ensure the referenced ModelEndpoint exists before running tasks.

If you want Anthropic routing instead of OpenAI routing, apply:

go run ./cmd/orlojctl apply -f examples/resources/model-endpoints/anthropic_default.yaml
go run ./cmd/orlojctl apply -f examples/resources/secrets/anthropic_api_key.yaml

If you want Azure OpenAI routing, apply:

go run ./cmd/orlojctl apply -f examples/resources/model-endpoints/azure_openai_default.yaml
go run ./cmd/orlojctl apply -f examples/resources/secrets/azure_openai_api_key.yaml

If you want local Ollama routing, apply:

go run ./cmd/orlojctl apply -f examples/resources/model-endpoints/ollama_default.yaml

Cyclical Manager/Research Loop (A <-> B)

This scenario shows explicit bidirectional handoffs (manager-agent -> research-agent -> manager-agent) with a bounded turn count (Task.spec.max_turns).

go run ./cmd/orlojctl apply -f examples/resources/agents/manager_agent.yaml
go run ./cmd/orlojctl apply -f examples/resources/agents/research_agent.yaml
go run ./cmd/orlojctl apply -f examples/resources/agent-systems/manager_research_loop_system.yaml
go run ./cmd/orlojctl apply -f examples/resources/tasks/manager_research_loop_task.yaml

Run workers/controller in task-execution-mode=message-driven with runtime inbox consumers enabled (--agent-message-consume) so inter-agent handoff messages are processed.

Governance UI Scenario (Denied)

This scenario intentionally denies one tool call so governance chips appear in runtime timelines.

go run ./cmd/orlojctl apply -f examples/resources/agent-roles/analyst_role.yaml
go run ./cmd/orlojctl apply -f examples/resources/tool-permissions/web_search_invoke_permission.yaml
go run ./cmd/orlojctl apply -f examples/resources/tool-permissions/vector_db_invoke_permission.yaml
go run ./cmd/orlojctl apply -f examples/resources/agents/research_agent_governed.yaml
go run ./cmd/orlojctl apply -f examples/resources/agent-systems/report_system_governed.yaml
go run ./cmd/orlojctl apply -f examples/resources/tasks/weekly_report_governed_task.yaml

Governance UI Scenario (Allowed)

Adds the missing role permission for vector DB so the run can proceed.

go run ./cmd/orlojctl apply -f examples/resources/agent-roles/vector_reader_role.yaml
go run ./cmd/orlojctl apply -f examples/resources/agents/research_agent_governed_allow.yaml
go run ./cmd/orlojctl apply -f examples/resources/agent-systems/report_system_governed_allow.yaml
go run ./cmd/orlojctl apply -f examples/resources/tasks/weekly_report_governed_allow_task.yaml

Load Test Retry-Stress Scenario Resources

These manifests back the orloj-loadtest retry-stress injection mode.

go run ./cmd/orlojctl apply -f examples/resources/agents/loadtest_timeout_agent.yaml
go run ./cmd/orlojctl apply -f examples/resources/agent-systems/loadtest_timeout_system.yaml

WASM Tool Reference Module

Apply the wasm tool resource:

go run ./cmd/orlojctl apply -f examples/resources/tools/wasm-reference/wasm_echo_tool.yaml

Run the reference guest module directly:

wasmtime run --invoke run examples/resources/tools/wasm-reference/echo_guest.wat

Use wasm isolation mode in worker/control-plane binaries and point module path at this file:

go run ./cmd/orlojworker \
  --tool-isolation-backend=wasm \
  --tool-wasm-module="$(pwd)/examples/resources/tools/wasm-reference/echo_guest.wat" \
  --tool-wasm-entrypoint=run