Technology Deep Dive — Section 13

Agents Built
For Your Organisation

Beyond the five core AI team members, configure bespoke designer agents for any domain-specific job your organisation needs done. No code. No deployment. Just configuration.

9 configuration knobs No code required Knowledge compounds with use

Every Organisation Has Unique Needs

Generic AI platforms give you one-size-fits-all

  • Same behaviour for every task
  • No domain-specific knowledge
  • No organisation-specific constraints
  • You adapt to the tool

EnGenAI bespoke agents are configured for you

  • Identity tailored to your domain
  • Constraints that match your policies
  • Connected to your specific systems
  • The tool adapts to you

9 Configuration Knobs

Configuring an agent means setting nine parameters. All nine are set through the platform API — Keith handles this on your behalf when you describe what you need.

01

Identity

Live

A natural-language prompt that defines who the agent is, what domain it operates in, and what topics it must cover. The LLM uses this to guide every conversation — no hard-coded question flows needed.

"You are a Legal Contract Reviewer. Your role is to identify non-standard clauses, flag risk areas, and summarise obligations. Do not advise on jurisdiction-specific law."

02

LLM

Live

Choose which AI model powers the agent and from which provider. Use a frontier model for complex reasoning tasks, an efficient model for simple extraction. Cost and quality, tuned per agent.

claude-opus-4-6 for contract analysis · claude-haiku-4-5 for data extraction

03

Steering

Live

Three-layer behavioral constraints: platform floor (inviolable), org floor (admin-set), and agent config. Lower layers can only add restrictions — never remove them. Guarantees agents cannot exceed their mandate.

"no_implementation_suggestions": true · "require_human_approval_on_write": true

04

Tools

Live

Bind vetted platform capabilities to the agent. MCP-registered servers (registered via the MCP Server Registry) appear here alongside built-in platform capabilities. Each tool has a permission tier (READ / WRITE / EXECUTE / DESTRUCTIVE) and a minimum autonomy level. Agents can only use what they have been explicitly given.

web_search (READ) · write_database (WRITE, min autonomy 1) · deploy_staging (EXECUTE)

05

Autonomy

Live

Five levels from fully human-controlled (0) to fully autonomous (4). Level 4 is reserved for internal EnGenAI agents only — customers can configure agents up to level 3. Plan tier gates maximum autonomy.

Level 1: AI suggests, human confirms each action · Level 2: AI acts, human reviews results

06

Connectors

Live

Connect the agent to your organisation's external systems — BI tools, project trackers, databases, any REST API — through a unified 4-layer permission model. Credentials live in GCP Secret Manager, never in the database.

Tableau (visualisation) · Jira (project tracking) · Snowflake (data warehouse) · Any REST API

07

Knowledge

Live

An org-scoped vector store (Milvus) that grows with every session. The agent retrieves relevant prior interactions before responding — knowledge compounds. Data is strictly partitioned: one organisation's knowledge never leaks to another.

"Previously, this stakeholder mentioned the refresh cycle must be daily — reflecting that context now."

08

Interface

Live

How users interact with the agent. Start with a web chat interface with topic-progress tracking. Extend to API invocation for programmatic use, or email mailbox for async stakeholder engagement.

Chat (web) · API (programmatic) · Email mailbox (async) — coming later

09

Handoff

Live

What happens when the agent completes its task. Route the output to a human reviewer by role (BA, architect, legal), set an SLA, and define what happens on approval or rejection. Powered by the Human Agent Integration system — see the dedicated deep dive.

On completion → assign to BA for review (SLA 24h) → on approve: export document → on reject: restart

Keith Configures It. You Just Describe It.

The CTO describes what they need. Keith — the CPO agent — makes all the API calls to provision the agent. No code change. No deployment needed.

1

You describe the agent to Keith

"I need an agent that gathers Tableau dashboard requirements from our data stakeholders"

2

Keith creates the Agent Template

POST /agent-templates — sets identity, LLM, autonomy, interface type

3

Keith binds the tools

POST /tool-bindings — web_search, write_database, and session management tools

4

Keith configures steering

Behavioural constraints: no implementation suggestions, GDPR-compliant, require approval on write

5

Keith connects external systems

POST /connector-bindings — Tableau workspace (read-only), HTTP endpoints for your data sources

6

Your agent is live

Immediately available in your workspace. Gets smarter with every session.

What Can You Build?

Any agent that needs domain expertise, organisation-specific constraints, and connections to your systems. These are just starting points.

Requirements Gathering

Captures stakeholder requirements for any software or data project. Guides a structured conversation, generates EARS-formatted requirements documents, and routes for human BA sign-off.

Chat interfaceDocument generationHuman approval gate

Legal Contract Review

Reads contracts, flags non-standard clauses, scores risk exposure, and produces a redline summary. Escalates to legal counsel for final sign-off before any contract is executed.

Document analysisRisk scoringEscalation workflow

BI Dashboard Consultant

Connects to your Tableau workspace, understands existing dashboard structures, and gathers requirements for new visualisations. Proposes metric hierarchies based on your data model.

Tableau connectorDomain knowledgeIterative sessions

Compliance Auditor

Reviews code changes and infrastructure configurations against your organisation's compliance policies. Flags violations, generates audit reports, and tracks remediation.

Policy enforcementAudit trailRemediation tracking
Platform Templates

Start From a Platform Template

Don't start from scratch. EnGenAI ships four pre-built Business Process Templates — fully configured agents ready to deploy into your organisation. All four have XML-structured steering enabled, delivering 15–30% fewer prompt tokens compared to Markdown format.

IDP

Intelligent Document Processing

Extracts structured data from unstructured documents — invoices, contracts, forms. Routes extracted data for human validation before downstream processing.

Document extractionHuman validation gateStructured output
REQ

Requirements Analysis

Guides structured requirements conversations with stakeholders. Generates EARS-formatted requirements documents and routes to a BA for sign-off.

EARS formatStakeholder conversationsBA approval gate
PAC

Predictive Analytics Consultant

Connects to your data warehouse, interprets trends, and proposes predictive models. Presents findings with human-readable rationale before any model is deployed.

Data warehouse connectorModel proposalsHuman rationale gate
DEV

Dev Request Handler

Receives, clarifies, and routes development requests. Decomposes work into actionable tickets, assigns estimates, and routes to the engineering team for prioritisation.

Request intakeTask decompositionEngineering routing

Platform templates are org-neutral — they appear in the template catalog for every organisation. Instantiate one via the API and it inherits your org's Steering Floor policies automatically. All four ship with XML-structured steering enabled.

Agents That Learn With Use

Every session adds to the agent's knowledge store — an org-scoped vector database that is strictly private to your organisation. The more your team uses a custom agent, the better it becomes at understanding your organisation's specific context, terminology, and patterns.

Session 1
Asks all foundational questions
Session 10
Recalls your preferences and patterns
Session 50+
Deep domain expertise about your organisation

Knowledge is strictly org-scoped. One organisation's data never influences another's agents.

Platform Status

Backend Live

Agent templates, tool catalog, autonomy levels, steering resolver, and human agent system — all deployed and serving requests.

Frontend Coming

Visual Agent Builder Wizard, tool catalog browser, and human task queue UI — shipping soon.

GOOGLE A2A PROTOCOL

Every custom agent exposes an A2A-compatible Agent Card endpoint for standardised inter-agent discovery.

AUTONOMY GATING

Autonomy levels above Level 2 require Pro or Enterprise plans. Starter customers build agents with human approval at every step.

See also: Human Agent Integration and Billing

Next: Transparent Billing

Custom agents are powerful. See how every token they use is metered, billed, and visible — with pre-paid credits, auto top-up, and hard-stop protection.