What You Get
- An agent with its prompt, tools, and integrations built and tested
- A golden test set and an offline eval harness you can re-run
- Prompt-injection defense and scoped, audited tool permissions
- Cost caps, rate limits, and runaway protection
- A production runbook and an incident SLA
Packages
Simple
from $10,000
A single-purpose agent — one job, a few tools, in production.
- Prompt + tools + integration
- Golden test set + offline evals
- Cost caps + production deploy
RAG Workflow
$18,000–$25,000
An agent grounded in your knowledge with retrieval and workflow logic.
- RAG corpus + retrieval evals
- Multi-step workflow + tool-use
- Injection defense + audit logging
Multi-Agent / Enterprise
$30,000–$50,000+ · Pilot-to-Production from $165,000
Multiple coordinated agents, or a governed 90-day pilot-to-production program with ISO-42001-aligned documentation.
- Multi-agent orchestration + observability
- Behind your IAM/SSO with audit logging
- Governance package (AI inventory, risk, monitoring) on the enterprise program
In Scope
Discovery
Weeks 1–2- Use-case selection and impact sizing
- Data inventory, sensitivity classification, access mapping
- Eval-harness design — golden set, synthetic edge cases, red-team prompts
- Architecture: model choice, RAG vs tool-use, deployment surface
Build
Weeks 3–8- Agent on AWS Strands, model-independent via a gateway
- MCP integration layer with audit logging and scoped permissions
- RAG indexing with retrieval evals, where applicable
- Observability: traces, evals, cost telemetry, anomaly alerts
Harden & Ship
Weeks 9–10- Eval pass against golden and red-team sets
- Cost ceilings, rate limits, human-in-the-loop escalation
- Production deploy behind your IAM, runbook, knowledge transfer
How It Works
01
A scoping call, then a Blueprint Sprint locks scope, tier, and a fixed price.
02
A named senior pod builds against a fixed timeline — no hourly billing.
03
Production deploy behind your auth, full IP, optional Run & Care after.
What “Done” Means
Your agent runs in production behind your auth, with eval scores you can see.
Why Codenovai
We build agents behind evals, guardrails, and audit logs — the control system that keeps them stable — on the same framework and patterns we run in our own products.
FAQ
- Is the model locked in, or can we swap providers?
- We route through a model gateway, so the agent calls a provider as a string and you can swap between Claude, GPT, Gemini, or a future model with a config change. Your prompts, tools, and evals stay portable — we avoid SDK patterns that would tie you to one foundation model.
- What stops the agent from doing something harmful or expensive?
- Scoped, audited tool permissions, prompt-injection defense, cost caps, rate limits, and human-in-the-loop escalation for ambiguous cases — all built in, not bolted on. You can see traces and eval scores from day one of production.
- How do we know it's actually working in production?
- Three numbers, reviewed weekly: task success against the golden set, average cost per resolved task, and human-escalation rate. If any drifts past the agreed threshold, we pause and root-cause before rolling further.
- What is the Pilot-to-Production program?
- A fixed-scope 90-day enterprise program: one agent into production behind your IAM with an eval harness, observability, and an ISO-42001-aligned governance package. Your team learns the operating model on the way through, so the next agent costs less.