AI Agents
Discovery Express

Platform

The layer where your agents run.

Five layers between your systems of record and the channel where the work happens. In your cloud, on your rules.

Talk to an architectSee the agents

AWS by default, in your tenant. Azure and GCP on request. On-prem for voice.

Five thin slabs stacked like floors: aluminum at the base, frosted glass above, violet at the top
  1. 05Interfaces
  2. 04Govern
  3. 03Agents
  4. 02Context
  5. 01Connect
Your systems of record
  • CONTPAQi

Five layers

One runtime, five layers, nothing hidden.

Connect, Context, Agents, Govern and Interfaces: the five layers of the DaCodes enterprise agent platform, from your systems of record up to the channel.

Two speech bubbles, one of violet frosted glass and one of brushed aluminum
05

Interfaces

Voice, WhatsApp, Slack, Teams and the Console.

An aluminum shield behind a thin frosted glass plate
04

Govern

Evals, guardrails, audit trail and cost limits.

A frosted glass squircle with a small aluminum sphere set off-center
03

Agents

Multi-model runtime, orchestration and our own voice stack.

A stack of frosted glass cards, the top one tinted violet
02

Context

Entities, rules, documents and memory in one layer.

Two interlocking rings of frosted glass and brushed aluminum
01

Connect

Your systems of record, natively and over MCP.

Model optionality

Nothing you build is tied to one model.

Each task runs on the model that wins its eval. Routing lives in the runtime, so changing a model is a config change, not a rewrite.

Routing through Amazon Bedrock and OpenRouter. Model list reviewed Sep 2026.

  • Open weights
Model routing
Demo data
TaskRouted toEval
  • Invoice field extractionClaude96.4%
  • Ticket triageAmazon Nova94.8%
  • Voice turn, SpanishQwen3 32B, open weights92.7%
  • Weekly report summaryGemini95.3%
Change a model: one line of config, then the eval set runs again.

Infrastructure

The agents live next to your data, not ours.

In your AWS, Azure or Google Cloud account. Nothing is copied to a DaCodes environment.

Deployment map
Demo data
Your systems of record
Your AWS account · your VPCYour IAM roles

Agent runtime

Bedrock AgentCore · LangGraph

  • Amazon Bedrock

    models, routed

  • RDS

    Postgres · pgvector

  • OpenSearch

    hybrid search

  • S3

    documents, audit log

DaCodes pod

Deploys through your IAM. No data leaves the VPC.

Same layout on Azure and GCP. Voice can run on your own GPUs.

Where it runs

Your tenant, your IAM, your VPC.

We deploy inside your cloud account. Data never leaves your perimeter.

DefaultIn production

AWS

Bedrock, RDS, OpenSearch and S3 behind your IAM and your VPC.

AWS Advanced Tier Services Partner, Jun 2026

On requestReference architecture

Azure & GCP

Same architecture, same repo layout, same evals.

Reference architecture, Sep 2026

Own stackReference architecture

On-prem for voice

Your GPUs, when calls cannot leave the building.

Reference architecture, Sep 2026

What you keep

Your code, your cloud, your IP.

Everything lands in your accounts, including an exit plan we can execute in five days. We'd rather you stay because it works.

  • Source code in your repo
  • Infrastructure as code in your account
  • Prompts, evals and datasets, versioned
  • Runbooks and a two-day workshop
  • Exit plan executable in five days
Handover package · your-org/agents
Demo data

main · 8f3c2a1

  • agents/prompts, graphs, tools
  • infra/Terraform for your account
  • evals/312 cases, thresholds
  • datasets/versioned with each release
  • runbooks/on-call, workshop, exit plan
AWS account
4821 •••• 7730
owner: you
Exit plan
executable in 5 days

Talk to an architect

45 minutes with the people who run the platform. No slides.

We reply in under one business day.

Loading the form