Switchyard · Agentic AI + RAG Harness

Scenario presets start here

Stack framework · vectordb

Orchestrator

Vector DB

LLM provider

Embeddings

LLM parameters

Model name

Temperature

Max tokens

Stream tokens

System prompt

RAG pipeline tune me

Chunk size

Overlap

Top-K

Similarity threshold

Embedding dim

Reranker

Watch the Chunker node: chunk boundaries recompute live. Raise the threshold until retrieval misses to see the agent fall back.

Agent loop & memory

Pattern

Max iterations

Memory window (turns)

Long-term vector memory

On no-retrieval-hit

Multi-agent team

Multi-agent swaps the single loop for Planner → Researcher → Critic hand-offs.

Guardrails

Input filter

Blocked terms (comma-separated)

Output filter (redact PII)

Try a preset query containing a blocked term, or watch the output filter redact numbers in the answer.

Tools 3

Knowledge base 3 docs

Documents are chunked for real — edit bodies and watch chunk counts change.

Chaos engineering

Slow down

Fail next tool call

Inspector