Transitioning to enterprise software. Services live now. First product, RAG Studio, ships Q4 2026. See the roadmap →
Service · Cluster 1 · AI Dev Platform

Production-ready
AI agents in 6–10 weeks.

A fixed-scope engagement that takes a single high-value workflow from idea to a deployed, audited AI agent backed by a private RAG knowledge base. POPIA-compliant by architecture. Built on the same engine that powers our enterprise platform — and convertible to a subscription on day one of go-live.

One agent. One knowledge base. One deployment. No scope creep, no surprise invoices.
Built on LangGraph & LlamaIndex. Auditable execution traces and source-cited retrieval out of the box.
Hosted in AWS af-south-1. Or on your own infrastructure — AWS Cape Town, Azure South Africa North, or GCP Johannesburg.
POPIA documentation in the deliverable. Section 11, Section 19, and Section 72 controls evidenced.
POPIA-native
Fixed scope, fixed price
30-day post-delivery support
Convert to subscription on close
Starting at
R95,000
Delivery window
6–10 weeks
Default region
AWS af-south-1
Post-delivery
30-day support
01 · Outcomes

What you get on day one of go-live.

sonofgraig has shipped this engagement enough times to know exactly what makes it succeed: precise scope, an opinionated stack, and a delivery cadence that protects against the two biggest failure modes in enterprise AI — vague success criteria and out-of-control infrastructure costs.

Outcome 01
A working agent in your environment
Designed on the visual canvas, deployed behind your authentication, integrated with your tools (Slack, email, CRM, ticketing, databases). Connected to your own RAG knowledge base — not a generic foundation model.
5 business days · time to first internal demo
Outcome 02
A POPIA-compliant pipeline by architecture
PII detection and redaction before content reaches the embedding model. AES-256 at rest. Immutable query-level audit log. Data residency enforced at the VPC layer — not by contract, not by trust.
0 raw PII transmitted to external LLM APIs
Outcome 03
A package your CIO can sign off
Architecture diagrams, runbooks, the source code for the agent, the RAG ingestion config, the test corpus, and a POPIA Section 19 compliance dossier. Every deliverable lives in your repository, not ours.
100% of code, configs and docs handed over
02 · Who it's for

Built for South African enterprises that need to ship.

If your organisation has identified a single, high-value workflow where an AI agent could remove a bottleneck — and you have the documents, data, or systems to ground it in — this engagement is designed to take you from sign-off to production faster than an in-house build.

Persona A
Engineering & product leaders
Need a working AI agent in production fast, without diverting an internal team for six months. Want a stack that the team can take ownership of after handover — not a black box.
Persona B
Customer experience & ops teams
Have a clear support, sales, or operations workflow where AI could compress hours of repetitive work each day. Need an agent that respects POPIA and produces auditable decisions.
Persona C
CIOs & Information Officers
Cannot procure on hyperscaler-only AI services that route data offshore. Need a pilot that can be defended in a board pack and audited by the Information Regulator if asked.
03 · Use cases

Six workflows that ship in this engagement.

A single AI Agent Implementation engagement covers one workflow. The patterns below are the ones we have built repeatedly — each is a proven match for the 6–10 week scope. Bring your version of one of these, or bring something close.

Customer support
Tier-1 support agent grounded in your product knowledge
Answers product questions, drafts responses, logs tickets in Jira or Zendesk, escalates to a human with full conversation context. Built on your help-centre articles, runbooks, and policy documents.
SlackZendeskJiraRAG
Sales & CRM
Lead-qualification & account-research agent
Reads a new lead in HubSpot or Salesforce, enriches the account from the web, drafts a personalised first-touch email, and books a meeting through the rep's calendar. Human approval gate before send.
HubSpotSalesforceGmailBrowser
Legal & professional services
Contract & precedent research agent
Searches your firm's prior memos, contracts, and judgments. Returns source-cited summaries with paragraph-level references back to the original document. Document-level access permissions enforced.
SharePointRAGPII scrubbing
Finance & banking
Policy & procedure compliance agent
Helps your front-line staff answer customer questions consistent with current FAIS, FICA, and internal credit policy. Every response is logged for compliance review — immutable audit trail by default.
RAGAudit logApproval gates
Operations
Internal IT & helpdesk triage agent
Reads incoming tickets, classifies them, suggests likely resolutions from your historical resolution archive, and escalates anything outside the runbook to a human engineer with prefilled context.
JiraLinearSlackDatabase
Healthcare admin
Clinical-protocol lookup agent (admin only)
Answers operational and procedural questions for clinical and administrative staff — never patient-facing decisions. POPIA Section 26 special-PI safeguards enforced. Human oversight on every output.
RAGPOPIA s.26Human-in-loop
04 · Delivery cadence

Four phases, six to ten weeks, one signed deliverable.

Every phase ships a tangible artifact — a scoping doc, an architecture diagram, a working prototype, a production deployment. Nothing is left for "later". The cadence below is the standard plan; complex integrations or regulated industries can extend the production phase by up to two weeks.

Discovery & scoping
Workflow analysis · Success metrics · POPIA risk assessment
Week 1
Two on-site or remote workshops with your operating team and your Information Officer. We document the target workflow, identify the data sources, define the success metrics in measurable terms, and surface POPIA Section 11 lawful-basis questions before they become blockers.
Deliverables
Signed scoping document with measurable acceptance criteria
Data source map & processing-purpose statement
POPIA risk register with mitigations identified per source
Architecture & RAG ingestion
Knowledge base build · PII scrubbing · Retrieval evaluation
Weeks 2–3
We design the agent graph in LangGraph and build the RAG ingestion pipeline using LlamaIndex into a per-tenant Qdrant collection. The PII scrubber is configured for your document set and tested against synthetic SA personal-information patterns. Retrieval is evaluated using Ragas (faithfulness, answer relevancy) before any agent code is written.
Deliverables
Architecture diagram & LangGraph state machine specification
Ingested RAG knowledge base with retrieval evaluation report
PII scrubbing test report — ID nos, mobile, email, passport patterns
Agent build & tool integration
Tool wiring · Memory · Human-in-the-loop · Sandbox testing
Weeks 3–7
The agent is built on the React Flow canvas, then committed as code in your repository. Tools are integrated through Composio — Slack, Gmail, Jira, Linear, HubSpot, Salesforce, PostgreSQL, web browser, custom REST APIs. Memory architecture (short-term conversation, long-term vector, episodic) is configured. Human-in-the-loop approval gates are added for any consequential action. Iterative testing in the sandbox against the scenarios you supplied.
Deliverables
Source code for the agent — in your repository, your branch
Tool integration documentation per system & OAuth scopes used
Test corpus & passing run report from the sandbox
Deployment, handover & 30-day support
Production deploy · Runbooks · Compliance dossier · Knowledge transfer
Weeks 7–10
The agent is deployed in production on your infrastructure (AWS af-south-1 by default, or Azure South Africa North or GCP Johannesburg). Observability (Prometheus, Sentry, Langfuse) is wired in. Kill switches are tested. Runbooks are written for rollout, rollback, model swap, and incident response. Two knowledge-transfer sessions are run with your team. The 30-day post-delivery support window starts on go-live — we resolve issues, tune retrieval, and tweak prompts.
Deliverables
Production deployment with observability and kill-switch verified
Runbooks & on-call procedure aligned to your incident process
POPIA compliance dossier — ss.11, 19, 72 evidence pack
30 days of priority support and tuning included from go-live
05 · Architecture

The same five layers as our enterprise platform.

Every agent we deliver is built on the architecture that powers sonofgraig's platform. That matters for two reasons: first, the security and POPIA controls are inherited and not bolted on. Second, when you choose to convert to a platform subscription, the agent moves over without rework.

L01
Edge & ingress
Cloudflare WAF, bot management, rate limiting, and DDoS mitigation in front of every public endpoint. TLS 1.3 terminated at the AWS Application Load Balancer. JWT validated, organisation context propagated downstream.
Cloudflare WAF TLS 1.3 AWS ALB
L02
Agent runtime & AI gateway
LangGraph executes the agent as a stateful directed graph — nodes for LLM calls, tool calls, and human-approval interrupts; conditional edges for routing. The PII scrubber runs synchronously before any external LLM call. LiteLLM normalises calls across Anthropic Claude, with Gemini available on Growth and above.
LangGraph LiteLLM PII scrubber Claude
L03
RAG knowledge base
LlamaIndex orchestrates ingestion, chunking, embedding, and retrieval. Qdrant stores embeddings in a per-tenant collection. Unstructured.io handles complex document parsing for PDF, Word, and PowerPoint. Ragas evaluation is run on every change to the index.
LlamaIndex Qdrant Unstructured.io Ragas
L04
Tools & integrations
Composio provides 200+ pre-built tool integrations: Slack, Gmail, Outlook, GitHub, Jira, Linear, HubSpot, Salesforce, PostgreSQL, MySQL, web browser via Playwright, sandboxed code execution. Custom REST APIs are added through Composio's connector builder.
Composio Slack / Gmail Jira / Linear Playwright
L05
Infrastructure & observability
Production runs on AWS EKS in af-south-1 by default. PostgreSQL via Supabase with row-level security per organisation. Redis (Upstash) for caching and Celery task queues. Terraform for everything. Prometheus, Grafana, Sentry, and Langfuse cover infrastructure metrics, error tracking, and LLM observability.
AWS af-south-1 EKS Kubernetes Supabase RLS Terraform Prometheus + Grafana Sentry Langfuse
06 · Scope

Exactly what's in. Exactly what's not.

Fixed-scope means we have to be explicit about boundaries. The lists below are the standard inclusions and exclusions for the R95,000 starting price. Anything in the right column can be quoted as a separate engagement — or rolled into a sonofgraig platform subscription on conversion.

Included
In the fixed-scope engagement
  • Discovery, scoping, and POPIA risk assessment
  • Architecture design and LangGraph state machine specification
  • RAG ingestion pipeline for one knowledge base (up to 5,000 documents)
  • PII scrubber configured and tested for SA personal-information patterns
  • Single-agent build with up to 6 tool integrations via Composio
  • Memory architecture (short-term, long-term, episodic)
  • Human-in-the-loop approval gates for consequential actions
  • Production deployment on AWS af-south-1 (your account or ours)
  • Observability stack — Prometheus, Sentry, Langfuse
  • Runbooks: rollout, rollback, model swap, incident response
  • POPIA compliance dossier — ss.11, 19, 72 evidence
  • Two knowledge-transfer sessions with your team
  • 30 days of priority support from go-live
Out of scope
Quoted separately
  • More than one agent or more than one knowledge base
  • Custom model fine-tuning — covered by Fine-Tuning Ops product
  • Multi-agent supervisor-worker orchestration
  • Voice agent or telephony integration
  • Source-system schema changes or data engineering work
  • End-user UI design beyond a basic embeddable chat widget
  • Penetration testing of integrated source systems
  • POPIA Information Officer outsourcing — you retain that role
  • LLM token costs — metered to your provider account
  • Long-running operational support beyond the 30-day window
  • Bias auditing and XAI — covered by the Governance Hub product
07 · Technology stack

An opinionated stack. Open source where it matters.

We do not invent the engine; we invest where the value is. Every component below is production-grade open source or a SaaS service we consciously chose not to rebuild. You inherit the same engineering decisions our platform was built on — and you keep the source code.

Component
Category
Role in your agent
LangGraph
Agent runtime
Stateful graph-based orchestration. Powers every agent decision, tool call, and human-in-the-loop interrupt. MIT licensed.
LlamaIndex
RAG framework
Document ingestion, chunking, embedding, indexing, and retrieval over your private knowledge base.
Qdrant
Vector database
Self-hosted in af-south-1. One isolated, encrypted collection per knowledge base for POPIA data residency.
Composio
Tool integrations
200+ pre-built integrations — Slack, Gmail, Outlook, GitHub, Jira, Linear, HubSpot, Salesforce, databases, browsers.
LiteLLM
AI gateway
Single interface to Anthropic Claude. Handles routing, fallbacks, and token-usage tracking.
Anthropic Claude
LLM
Default reasoning model. PII-scrubbed payloads only. Gemini available on Growth tier and above on conversion.
PostgreSQL (Supabase)
Relational data
Stores agent configs, conversation history, and the immutable audit log. Row-level security per organisation.
Redis (Upstash)
Cache & queue
Conversation cache and async task queue for ingestion jobs.
AWS EKS · Terraform
Infrastructure
All workloads containerised on EKS in af-south-1. Every infrastructure change reviewable as Terraform code.
Cloudflare
Edge security
WAF, DDoS mitigation, bot protection, rate limiting per IP and per organisation.
Prometheus + Grafana
Metrics
Infrastructure observability — CPU, latency, request rate, error rate, SLO tracking.
Sentry
Error tracking
Frontend and backend error capture, performance monitoring, session replay.
Langfuse (self-hosted)
LLM observability
Prompt traces, token costs per agent run, evaluation scores, conversation logs.
Ragas
RAG evaluation
Faithfulness, answer relevancy, context precision — run on every change to the knowledge base index.
08 · Pricing

One number. No hourly surprises.

sonofgraig service projects are deliberately simple to procure. The price is the price. Scope is fixed before contracting. Variations are quoted in writing and signed before any additional work is performed.

AI Agent Implementation
Fixed-scope engagement
R95,000 ZAR
Starting price. Final figure depends on the source-system count, document volume, and integration complexity surfaced during scoping.
Single payment. 50% on contract signature, 50% on go-live.
6–10 weeks. Standard delivery window. Regulated industries may add up to 2 weeks.
30 days of post-delivery support. Priority response, retrieval tuning, prompt iteration.
POPIA documentation included. No separate compliance bill at the end.
Book a scoping call
What sits outside the engagement price
LLM token consumption Your provider bill
Cloud infrastructure costs (compute, storage) Pass-through
Additional knowledge bases beyond the first +R20K each
Additional agent beyond the first +R45K each
Continued support after 30-day window From R12K/mo
Bias audit / XAI report Governance Hub
Custom fine-tuning Fine-Tuning Ops
Convert to platform on close. The agent and knowledge base move directly onto a sonofgraig subscription with no rebuild. Your first three months on the corresponding plan are credited against the implementation fee — effectively R7,500 of free platform usage on Growth tier.
09 · Service to platform

Same engine. Same code. Bigger plan.

Most engagements convert into a platform subscription on close. The decision is operational, not technical — the agent already runs on platform components. The table below sets out which capabilities belong to the service engagement and which unlock when you move to a platform subscription.

Capability
Service engagement
Platform subscription
One agent + one knowledge base
Included
Unlimited on Growth+
Source code in your repository
You own it
Builder canvas + your code
Managed hosting on af-south-1
30 days
Always-on
SLA
Best effort
99.9% (Growth) · 99.95% (Enterprise)
Continuous tuning & retrieval evaluation
First 30 days
Ongoing
Governance Hub (bias audits, XAI, AI risk register)
Quoted separately
Available on Growth+
Fine-tuning
Out of scope
Fine-Tuning Ops on Enterprise
BYOK encryption
Not included
Enterprise tier
Multi-agent supervisor — worker patterns
Out of scope
Available on Growth+
Pricing
From R95,000 once
From R4,999/month
10 · Frequently asked

Questions procurement, legal and engineering ask.

If your team is preparing for a vendor review or a board sign-off, the answers below cover most of what gets raised. Anything else, your account team can route to engineering directly.

Is the R95,000 fixed, or just a starting figure?
It is the starting price for the standard scope — one agent, one knowledge base of up to 5,000 documents, six tool integrations, deployment in af-south-1, 30 days of support. Your final fixed price is confirmed at the end of the scoping phase, before any contract is signed. Once signed, the price does not move unless you formally request additional scope, which is quoted in writing and re-signed before work continues.
What does the LLM token bill look like in practice?
Token costs sit on your provider account, not ours, so you have full visibility. We model expected token usage during scoping based on the conversation pattern, the average context size, and expected daily volume. As an order-of-magnitude reference, a tier-1 customer-support agent handling a few thousand conversations a month typically lands between R3,000 and R8,000 in monthly LLM cost on Anthropic Claude. This is independent of the implementation fee and can be capped via budget alerts.
Where does our data physically live during the engagement?
By default, all production data sits in AWS af-south-1 (Cape Town, South Africa) — the same default our enterprise platform uses. We can also deliver into Azure South Africa North or GCP Johannesburg if your group security policy dictates a specific provider. PII-scrubbed payloads are the only data ever sent to an external LLM provider, and that boundary is enforced by middleware, not policy.
Who owns the source code at handover?
You do. The source code, infrastructure-as-code (Terraform), CI/CD configuration, runbooks, the test corpus, and the POPIA documentation are committed to your repositories during the engagement — not at the end. sonofgraig retains no proprietary lock-ins on the agent or its knowledge base. If you choose not to convert to a platform subscription, the agent runs on standard open-source components your team can maintain.
Can the engagement be extended if scope changes?
Yes — through a formal change request. Common extensions are an additional knowledge base (typically +R20K), an additional agent (+R45K), or an additional integration beyond the standard six (priced per integration). Change requests are quoted in writing, signed by both parties, and only billed once accepted. We will never present an unexpected line item at the end of the engagement.
What does the 30-day post-delivery support cover?
Priority response on issues, retrieval tuning when answers drift, prompt iteration based on real-world usage, and minor adjustments to tool configurations. It does not cover net-new features, additional integrations, or operational on-call — those are quoted as continued support from R12K/month or are included on a platform subscription.
Do you sign Data Processing Agreements?
Yes. sonofgraig has a pre-signed Data Processing Agreement covering processing activities, lawful basis, security controls, sub-processors, and transfer mechanisms. It is available for download from our trust centre at /dpa and your legal team can mark up departures from the standard text during contracting.
Do we need to be on a platform subscription before or after?
No subscription is required to engage the service — that is the point. Many customers buy this engagement specifically to validate the technology and the supplier before committing to recurring spend. If you do convert at close, the first three months on the corresponding plan are credited against the implementation fee, which effectively returns R7,500 of free Growth-tier usage.
Are sonofgraig B-BBEE certified and CIPC registered?
Yes — sonofgraig is B-BBEE certified and CIPC registered. B-BBEE spend certificates are issued per invoice. All commercial documentation is available to your procurement team for supplier on-boarding.
Ready to scope

Book a 30-minute scoping call.

A senior solutions engineer joins, we step through the workflow you have in mind, identify whether it fits the standard scope, and confirm what your final fixed price will be. No commitment until contract signature.