Formulate RAG

Private search, per tenant. Not six months of pipeline work.

Q: How does Formulate RAG relate to Nyas and Anchor?

Formulate RAG is the developer-facing ingest and search API for the retrieval engine that runs on Nyas — managed Postgres with pgvector. It is the same retrieval logic that gives Anchor its agent memory, exposed directly to your product so you do not have to provision a separate vector database.

Q: What does multi-tenant RAG mean?

Multi-tenant RAG gives each customer or department an isolated knowledge namespace while operating one shared product deployment.

Q: Can Formulate RAG run inside my cloud?

Yes. The BYOC deployment runs retrieval inside your AWS, Azure, or GCP environment so your source data remains in your cloud account.

Q: Which document formats can I ingest?

The current ingestion pipeline supports PDF, DOCX, TXT, CSV, and HTML documents.

Q: How does my application call the RAG service?

Applications use an ingest endpoint to add documents and a search endpoint to retrieve relevant context, passing the tenant identifier with each request.

Q: Can I use Formulate RAG for an embedded product feature?

Yes. It is designed for SaaS products and enterprise applications that need private search or question answering for many isolated customers or departments.

Formulate RAG turns the documents your customers or departments already have into grounded, cited answers — chunking, embeddings, ranking, and per-tenant isolation included. Two API calls: one to ingest, one to search. Hosted on our infrastructure or deployed inside your own cloud.

Start 3-month pilot Book a technical demo

Runs on Nyas (Postgres + pgvector) — the same retrieval engine that gives Anchor its agent memory. No separate vector database to provision.

<300ms

P99 retrieval latency

API calls to integrate

72h

from agreement to live

∞

tenants, one deployment

The build-vs-buy reality

Building RAG in-house costs more than it looks.

Every team underestimates it. The chunking is easy. Multi-tenancy, re-ranking, latency at scale, embedding drift, security audits — that's where the 6 months go.

Engineering cost

$150–200K

2 engineers × 6 months to build + maintain

Time to market

6 months

before your first customer can search a document

Ongoing cost

$50K+/yr

maintenance, upgrades, security patches

Formulate RAG: flat monthly license. Live in 72 hours. No engineering headcount. No maintenance burden. Scale from 10 tenants to 10,000 without changing a line of code.

How it works

Document in. Answer out. Every tenant isolated.

📄

Documents

PDF · DOCX · TXT · CSV

▶

⬆

/ingest

POST with tenant ID

▶

🗄

Nyas

pgvector · S3

▶

🔍

/search

<300ms P99

▶

✅

Results

chunks + scores + source

Each tenant gets a fully isolated namespace

Tenant AAcme Corp

📄policy.pdf

📄handbook.docx

ISOLATED NAMESPACE

Tenant BBeta Inc

📄contracts.pdf

📄sops.txt

ISOLATED NAMESPACE

Tenant CGamma Ltd

📄specs.pdf

📄wiki.html

ISOLATED NAMESPACE

🔒Tenant A's searches never touch Tenant B's data — isolated at the pgvector layer, not via filters.

Who it's for

Two customer types. Two deployment models.

B2B2CSaaS Platforms

You have customers. They need their own knowledge base.

The problem

Your customers want document search, Q&A on their own data, or a private knowledge base — inside your product. Building multi-tenant RAG takes 2 engineers × 6 months.

What you get

RAG becomes a silent feature inside your product. Each customer gets a fully isolated knowledge base. You deploy in your own cloud, data never leaves your VPC. Your customers never know it's Formulate.

Used by

Messaging platformsCRM toolsSupport softwareChatbot productsDocument management

BYOC — $5,500/mo · Unlimited tenants

B2BEnterprises

You have departments. Each one needs its own RAG.

The problem

One org, many teams — HR, Legal, Finance, Clinical. Each needs to search their own documents without seeing each other's data. No DevOps team to maintain infrastructure.

What you get

Call our hosted API. Each department gets a fully isolated namespace. No infra work, no DevOps. Pass the tenant ID in the header — isolation is handled at the database layer.

Used by

Hospitals & pharmaBanks & NBFCsInsurance firmsUniversitiesLarge conglomerates

Hosted API — from $800/mo

Compliance & Security

Built for regulated industries from day one.

Every enterprise RAG deployment runs a compliance gauntlet — DPDP, HIPAA, RBI, SEBI, SOC2. Formulate RAG was designed with these requirements baked in, not bolted on.

India DPDP Act

Data Principal and Fiduciary obligations met. Data stays within India on BYOC deployments.

HIPAA Ready

Healthcare-grade data handling. BAA available. Air-gapped deployment for regulated clinical data.

RBI / SEBI Guidelines

Financial data never co-mingled across tenants. Native isolation satisfies regulatory audit requirements.

SOC 2 Type II

Evidence collection automated. Audit-ready posture. Security controls documented and tested.

Data Sovereignty

BYOC: retrieval never leaves your cloud. Generation via Bedrock or Azure OpenAI stays in your account.

Tenant Isolation

Native isolation at the pgvector layer — not a filter, not access control. Architecturally impossible for tenants to cross-query.

True data sovereignty in BYOC mode: Retrieval never leaves your cloud. For generation, use AWS Bedrock or Azure OpenAI — both keep LLM calls within your cloud account. Bringing your own OpenAI key routes generation through OpenAI's servers — not ours.

Integration

Two API calls. That's the whole integration.

POST to /ingest to upload documents. POST to /search to retrieve. Pass your tenant ID in the header. Done.

Step 1 — Ingest documents

api.formulateai.in/v1/ingest201 Created

POST /v1/ingest
Authorization: Bearer fai_live_51NkQ2xR8m
X-Tenant-Id: acme-corp
Content-Type: multipart/form-data

file: Q3_Enterprise_Pricing_Guide.pdf  (2.4 MB)

→ 201 Created
{
  "document_id": "doc_8f2a1c90e4",
  "tenant_id": "acme-corp",
  "status": "processing",
  "pages": 34,
  "estimated_chunks": 212,
  "webhook_url": "https://acme.app/webhooks/formulate"
}

Step 2 — Search at query time

api.formulateai.in/v1/search200 OK · 214ms

POST /v1/search
Authorization: Bearer fai_live_51NkQ2xR8m
X-Tenant-Id: acme-corp
Content-Type: application/json

{
  "query": "What is the refund window for annual enterprise contracts?",
  "top_k": 3
}

→ 200 OK · 214ms
{
  "results": [
    {
      "content": "Annual enterprise contracts include a 30-day...",
      "score": 0.93,
      "source": "Q3_Enterprise_Pricing_Guide.pdf",
      "page": 12
    },
    {
      "content": "Refunds on mid-term downgrades are prorated to...",
      "score": 0.81,
      "source": "MSA_v4_2025.docx",
      "page": 6
    }
  ]
}

PDF · DOCX · TXT · CSV · HTMLAsync ingestion + webhooksSemantic search (pgvector)Hybrid BM25 + vectorTop-k with scores + sourcesLLM-agnostic (returns chunks)

Deployment

Your cloud or ours. Your choice.

Hosted API

We host. You call.

Formulate AI runs the full RAG stack on our infrastructure. You make two API calls — /ingest and /search. We handle availability, scaling, and ops.

No infra work on your side
SLA-backed 99.5% uptime
Live in 72 hours
Monthly flat pricing

Best for enterprises getting started fast

Your Cloud (BYOC)

Your cloud. Our software.

Deploy Formulate RAG inside your own AWS, Azure, or GCP account. You own the data, the infra, and the compute. We provide the software and the deployment package.

Data never leaves your VPC
Bedrock / Azure OpenAI for generation
CloudFormation + Helm chart included
Software license — you pay cloud bill

Best for platforms with data sovereignty requirements

Why Formulate RAG

Everything you'd otherwise spend six months building.

Multi-tenancy, isolation, and deployment flexibility come standard — not add-ons you negotiate for later.

Multi-tenant isolation out of the box — not a filter, a separate namespace per tenant

Deploy hosted on our infrastructure, or inside your own AWS, Azure, or GCP account

LLM-agnostic — we return ranked chunks, you generate the answer with any provider

Flat monthly pricing per tenant tier, not metered per document or per query

Live in 72 hours from signed agreement to your first tenant querying real documents

Zero infrastructure work on your side — no vector database to size, patch, or scale

Pricing

Flat pricing. No per-query surprises.

One-time setup fee $2,500 covers integration support, developer onboarding, and 30-day hypercare. Annual billing saves 15%.

Starter

$800/ month

Up to 100 tenants

$8 / tenant

REST API (ingest + search)
Tenant-isolated namespaces
PDF, DOCX, TXT, CSV, HTML
Async ingestion + webhooks
Email support

Get started

Growth

$2,500/ month

Up to 500 tenants

$5 / tenant

Everything in Starter
Per-tenant usage dashboard
Query logs + analytics
Slack support channel
Response < 4 hours

Get started

Scale

$4,000/ month

Up to 1,000 tenants

$4 / tenant

Everything in Growth
Dedicated onboarding engineer
30-day hypercare
99.5% uptime SLA
Annual: $40,800 (15% off)

Get started

BYOC

$5,500/ month

Unlimited tenants

You pay your cloud bill

Deploy in your AWS/Azure/GCP
Bedrock / Azure OpenAI ready
CloudFormation / Helm chart
Full data sovereignty
One-time setup: $3,000

Get started

Pilot offer: 3 months at $2,000/mo.

Up to 100 tenants. No commitment for full rollout. Prove it works in your product or org, then expand.

Start the pilot Book a 30-min walkthrough

contact@formulateai.in · Built in Bangalore · Data stays yours

Frequently asked questions

Clear answers before you start.

How does Formulate RAG relate to Nyas and Anchor?

Formulate RAG is the developer-facing ingest and search API for the retrieval engine that runs on Nyas — managed Postgres with pgvector. It is the same retrieval logic that gives Anchor its agent memory, exposed directly to your product so you do not have to provision a separate vector database.

What does multi-tenant RAG mean?

Multi-tenant RAG gives each customer or department an isolated knowledge namespace while operating one shared product deployment.

Can Formulate RAG run inside my cloud?

Yes. The BYOC deployment runs retrieval inside your AWS, Azure, or GCP environment so your source data remains in your cloud account.

Which document formats can I ingest?

The current ingestion pipeline supports PDF, DOCX, TXT, CSV, and HTML documents.

How does my application call the RAG service?

Applications use an ingest endpoint to add documents and a search endpoint to retrieve relevant context, passing the tenant identifier with each request.

Can I use Formulate RAG for an embedded product feature?

Yes. It is designed for SaaS products and enterprise applications that need private search or question answering for many isolated customers or departments.