pilot-knowledge-base-rag-setup（pilot 知识 base RAG 设置）

Knowledge Base (RAG) Setup

Deploy 4 agents: ingest, embed, index, and query.

Roles

Setup Procedure

Step 1: Ask the user which role and prefix.

Step 2: Install skills:

# ingest:
clawhub install pilot-s3-bridge pilot-share pilot-chunk-transfer pilot-cron
# embedder:
clawhub install pilot-task-parallel pilot-share pilot-metrics pilot-task-chain
# indexer:
clawhub install pilot-database-bridge pilot-share pilot-task-chain pilot-health
# query:
clawhub install pilot-api-gateway pilot-health pilot-load-balancer pilot-metrics

Step 3: Set hostname and write manifest to ~/.pilot/setups/knowledge-base-rag.json.

Step 4: Handshake along the pipeline: ingest↔embedder, embedder↔indexer, indexer↔query.

Manifest Templates Per Role

ingest

{
  "setup": "knowledge-base-rag", "role": "ingest", "role_name": "Document Ingestion",
  "hostname": "<prefix>-rag-ingest",
  "skills": {
    "pilot-s3-bridge": "Pull documents from S3 buckets.",
    "pilot-share": "Send document files to embedder.",
    "pilot-chunk-transfer": "Split large documents into chunks.",
    "pilot-cron": "Schedule periodic ingestion sweeps."
  },
  "data_flows": [{ "direction": "send", "peer": "<prefix>-rag-embedder", "port": 1001, "topic": "doc-ingested", "description": "Document chunks" }],
  "handshakes_needed": ["<prefix>-rag-embedder"]
}

embedder

{
  "setup": "knowledge-base-rag", "role": "embedder", "role_name": "Embedding Generator",
  "hostname": "<prefix>-rag-embedder",
  "skills": {
    "pilot-task-parallel": "Generate embeddings in parallel for throughput.",
    "pilot-share": "Receive docs from ingest, send embeddings to indexer.",
    "pilot-metrics": "Track embedding throughput and latency.",
    "pilot-task-chain": "Chain chunking and embedding steps."
  },
  "data_flows": [
    { "direction": "receive", "peer": "<prefix>-rag-ingest", "port": 1001, "topic": "doc-ingested", "description": "Document chunks" },
    { "direction": "send", "peer": "<prefix>-rag-indexer", "port": 1001, "topic": "embeddings-ready", "description": "Vector embeddings" }
  ],
  "handshakes_needed": ["<prefix>-rag-ingest", "<prefix>-rag-indexer"]
}

indexer

{
  "setup": "knowledge-base-rag", "role": "indexer", "role_name": "Vector Indexer",
  "hostname": "<prefix>-rag-indexer",
  "skills": {
    "pilot-database-bridge": "Write embeddings to vector database.",
    "pilot-share": "Receive embeddings from embedder.",
    "pilot-task-chain": "Chain indexing operations.",
    "pilot-health": "Monitor index health and query latency."
  },
  "data_flows": [
    { "direction": "receive", "peer": "<prefix>-rag-embedder", "port": 1001, "topic": "embeddings-ready", "description": "Vector embeddings" },
    { "direction": "receive", "peer": "<prefix>-rag-query", "port": 1001, "topic": "search-query", "description": "Search queries" },
    { "direction": "send", "peer": "<prefix>-rag-query", "port": 1001, "topic": "search-results", "description": "Ranked results" }
  ],
  "handshakes_needed": ["<prefix>-rag-embedder", "<prefix>-rag-query"]
}

query

{
  "setup": "knowledge-base-rag", "role": "query", "role_name": "Query Server",
  "hostname": "<prefix>-rag-query",
  "skills": {
    "pilot-api-gateway": "Accept search queries from external clients.",
    "pilot-health": "Monitor query endpoint health.",
    "pilot-load-balancer": "Distribute queries across indexer replicas.",
    "pilot-metrics": "Track QPS, latency, result quality."
  },
  "data_flows": [
    { "direction": "send", "peer": "<prefix>-rag-indexer", "port": 1001, "topic": "search-query", "description": "Search queries" },
    { "direction": "receive", "peer": "<prefix>-rag-indexer", "port": 1001, "topic": "search-results", "description": "Ranked results" }
  ],
  "handshakes_needed": ["<prefix>-rag-indexer"]
}

Data Flows

• ingest → embedder : document chunks (port 1001)
• embedder → indexer : vector embeddings (port 1001)
• query ↔ indexer : search queries and results (port 1001)

Workflow Example

# On ingest:
pilotctl --json send-file <prefix>-rag-embedder ./docs/guide.pdf
pilotctl --json publish <prefix>-rag-embedder doc-ingested '{"doc_id":"doc-42","chunks":24}'
# On embedder:
pilotctl --json publish <prefix>-rag-indexer embeddings-ready '{"doc_id":"doc-42","vectors":24,"dims":1536}'
# On query:
pilotctl --json task submit <prefix>-rag-indexer --task '{"query":"How does auth work?","top_k":5}'

Dependencies

Requires pilot-protocol skill, pilotctl binary, clawhub binary, and a running daemon.