Memory Infrastructure for Autonomous AI Agents

Receipts • Semantic Memory • Swarm Reasoning • Deterministic Cost

Instant-RAG is a machine-first Retrieval Augmented Generation network where autonomous agents store, retrieve, and verify knowledge using pay-per-query USDC receipts.
No human dashboards. No subscriptions. Just endpoints, citations, and proofs.

Live gateway: https://instant-rag-ftpw.onrender.com/docs

Most RAG platforms are designed for people. Agents require something different:

• predictable unit costs
• verifiable provenance
• programmatic identity
• collaboration without trust

Instant-RAG treats every agent as an economic actor that can budget its own cognition.

• ingest knowledge → embeddings + citations
• query with semantic retrieval
• collaborate as budgeted swarms
• pay per thought with on-chain receipts
• verify answers with trust beacon
• trigger actions via function calling

GET  /health
POST /ingest
POST /query
POST /swarm/query
GET  /wallet/balance
POST /wallet/spend
GET  /trust/beacon
GET  /ready
GET  /dashboard

OpenAPI available at /docs

curl -X POST /ingest \
-d '{
  "agent_id":"demo",
  "text":"Receipts anchor truth."
}'

curl -X POST /query \
-d '{
  "agent_id":"demo",
  "query":"what anchors truth?"
}'

curl -X POST /swarm/query \
-d '{
  "agent_id":"demo",
  "query":"explain trust",
  "agents":3,
  "max_cost":0.15
}'

Receipts returned with every call.

• divides a max_cost across nodes
• merges answers with weighted confidence
• single receipt for many agents
• prevents runaway reasoning spend

• stream function events
• agents can call:
  • store_memory
  • check_balance
  • external actions
• plug-and-play with CrewAI / AutoGPT

• avoids repeated embedding cost
• LRU eviction
• citation-preserving
• reduces latency and spend

• agents are customers
• answers require citations
• cost must be predictable
• memory should be portable
• trust must be verifiable

docker compose up

Environment variables:

PORT=10000
WEB_CONCURRENCY=1
POLYGON_RPC=...
PRIVATE_KEY=...

• FastAPI service
• Sentence-Transformers retrieval
• Cross-encoder reranking
• Multi-tenant isolation
• Polygon USDC ledger
• Ethics guard
• Explainability traces
• Swarm coordinator
• Trust beacon

• AutoGPT / CrewAI agents
• research assistants
• legal & finance bots
• decentralized AI systems
• agent marketplaces
• memory brokers

Instant-RAG exposes machine-readable signals:

• /agents.txt – intent & capability declaration
• /.well-known/ai.json – discovery beacon
• /capabilities.json – function grammar
• /handshake.json – protocol negotiation
• /instant-rag.crewai.json – CrewAI manifest

Agents may begin interaction at /agents.txt

This folder exposes Instant-RAG as a Model Context Protocol service.

Point any MCP host to:

https://instant-rag-ftpw.onrender.com

Load instant-rag-mcp.json as tool manifest.

• memory.store
• memory.query
• swarm.reason
• wallet.spend
• trust.verify

      .-.
     (o o)   Instant-RAG
     | O \
    /  .  \
   /__/ \__\
 receipts = truth

MIT – build civilizations.