Skip to content

README.md

Latest commit

 

History

History
126 lines (100 loc) · 5.51 KB

README.md

File metadata and controls

126 lines (100 loc) · 5.51 KB
Raw

OrKa Examples Catalog

This folder contains curated workflow examples demonstrating OrKa's core patterns: fork/join, routing, validation, memory operations, external tools, loops, and local LLM orchestration.

How to run

  • Start backend: orka-start
  • Run any example:
    • orka run examples/<workflow>.yml"your input"

Note: Private sets (folders prefixed with PRIVATE_) are excluded from this catalog.

🧭 GraphScout Agent Examples (NEW in v0.9.3)

graph_scout_basic.yml

  • Purpose: Demonstrate basic GraphScout intelligent routing with automatic path discovery
  • Pattern: GraphScout → dynamic agent selection → execution
  • Highlights:
    • Automatic path discovery and evaluation
    • Budget and safety controls
    • Intelligent decision making (commit_next/shortlist/no_path)
    • Works with search_agent, analyzer, response_builder

graph_scout_memory_aware.yml

  • Purpose: Advanced GraphScout with memory-aware routing and intelligent agent positioning
  • Pattern: GraphScout → memory readers (first) → processors → memory writers (last) → response builder
  • Highlights:
    • Memory-aware routing logic (readers first, writers last)
    • Multi-agent sequential execution
    • Semantic memory integration with knowledge base
    • Comprehensive analysis pipeline with memory persistence

🚀 Try GraphScout:

# Basic intelligent routing
orka run examples/graph_scout_basic.yml "What are the latest developments in quantum computing?"

# Memory-aware intelligent routing  
orka run examples/graph_scout_memory_aware.yml "Explain machine learning algorithms"

temporal_change_search_synthesis.yml

  • Purpose: detect a pivotal date, generate before/after queries, search both, and synthesize a timeline.
  • Pattern: fork → search → join → synthesize.
  • Highlights:
    • detect_change → date (DD/MM/YYYY)
    • generate_before_query / generate_after_query
    • search_before / search_after (DuckDuckGo)
    • synthesize_timeline_answer → final summary

memory_validation_routing_and_write.yml

  • Purpose: retrieval-first flow with validation-based routing, then write-back to memory.
  • Pattern: memory read → validation (binary) → router → (answer-from-memory | search→answer) → memory write.
  • Highlights: decay-aware memory read, sufficiency validator, short-term write with metadata.

memory_read_fork_join_router.yml

  • Purpose: read memories, process in parallel branches, merge, then route.
  • Pattern: memory read → fork processors → join → router.
  • Highlights: demonstrates multi-branch enrichment before decisions.

person_routing_with_search.yml

  • Purpose: detect if input relates to a person; on true, extract name and expand with search; else answer generically.
  • Pattern: binary decision → true-path enrichment (name → search → narrative) | false-path answer.
  • Highlights: openai-binary_6 + router_7 → openai-answer_10 → duckduckgo_8 → openai-answer_9.

cognitive_society_minimal_loop.yml

  • Purpose: minimal multi-agent society with role agents and agreement/quality loop.
  • Pattern: loop → roles → agreement score → stop at threshold.
  • Highlights: loop metadata, scoring strategies, convergence.

failover_search_and_validate.yml

  • Purpose: resilient answering using failover tree of agents.
  • Pattern: failover node → alternative strategies → eventual success.
  • Highlights: production-oriented fallback behavior.

conditional_search_fork_join.yaml

  • Purpose: run parallel search/enrichment branches and merge results.
  • Pattern: fork (multiple branches) → join → next steps.
  • Highlights: joined_results() helper for downstream templating.

multi_model_local_llm_evaluation.yml

  • Purpose: compare local LLMs/params for cost/latency/quality.
  • Pattern: multiple local LLM calls → metrics capture → comparative synthesis.
  • Highlights: local inference focus.

cognitive_loop_scoring_example.yml

  • Purpose: iterative refinement with loop scoring until threshold.
  • Pattern: loop node → sub-workflow per iteration → scoring strategies.
  • Highlights: regex/key/agent-based scoring and embedding fallback.

orka_framework_qa.yml

  • Purpose: simple end-to-end Q&A over prompts/docs.
  • Pattern: prompt → answer (optionally write memory if configured).
  • Use: quick smoke tests and demos.

routed_binary_memory_writer.yml

  • Purpose: use binary decisions to route to different memory writers.
  • Pattern: predicate → router → writer(s).
  • Highlights: structured metadata, TTL usage for categories.

validation_structuring_memory_pipeline.yml

  • Purpose: validate and structure outputs into a consistent schema before persisting.
  • Pattern: answer → validation/structuring → memory write (schema-aware).
  • Use: normalized artifacts for downstream consumption.

memory_types_short_long_test.yml

  • Purpose: observe TTL/decay across short- vs long-term entries.
  • Pattern: write with different types/TTLs → observe expiry/retention.
  • Use: validate decay configuration and cleanup.

Template helper functions

  • get_input(): safe access to user input.
  • joined_results(): merged outputs from fork/join.
  • get_agent_response('<agent_id>'): convenient access to prior responses.
  • safe_get(obj, key, default=''): guard against missing keys.

Tips

  • Keep the intended final node as the last OpenAI/Local LLM agent in the sequence.
  • Routers enqueue next agents; they do not produce user-facing output.
  • Prefer helper functions in prompts to avoid unresolved variables.
  • Ensure Redis/RedisStack backend and HNSW index are healthy for memory examples.