01
Governance
Teams adopt tools faster than leadership can set policies.
ITECS defines approved use cases, data rules, human review paths, and operating ownership before AI spreads.
Custom AI Agents
Custom AI Agents for Your Dallas Business
ITECS builds custom AI agents that connect securely to your proprietary data, tools, folders, codebases, and business systems — with human-in-the-loop controls, audit trails, and production support.
Custom AI does not mean one chatbot on one model. ITECS designs secure agents, project folders, CLI workflows, and retrieval systems around the way your team actually works. We can configure Claude projects, Codex workflows, ChatGPT workspaces, agentic RAG, human approval queues, and API-connected automations that retrieve the right context, take approved actions, and keep sensitive data controlled.
The Stakes
AI adoption fails without governance, security, and operations.
Most organizations do not need another disconnected AI experiment. They need a managed operating model that tells people what is approved, where data can go, which workflows deserve investment, and who owns reliability after launch.
ITECS Position
Managed Intelligence applies ITECS's 24-year managed IT and cybersecurity operating model to AI systems, prompts, agents, connectors, and employee adoption.
02
Security
Sensitive data moves into public prompts, unmanaged plugins, and disconnected workspaces.
ITECS brings managed-IT discipline to access, identity, data handling, and vendor selection.
03
ROI
AI experiments consume subscriptions and meetings without a measurable operating case.
ITECS starts with workflows, cost of delay, and measurable outcomes before recommending build work.
04
Integration
Useful pilots stall when they have to connect with Microsoft 365, CRM, service, or finance systems.
ITECS designs automation around the systems, permissions, approvals, and support model already in place.
Illustrative Planning Signal
73%modeled share of experiments that could stall without governance and ownershipA planning assumption to validate during discovery, not a reported client result or performance guarantee.
Your Team Has AI Ideas. The Hard Part Is Making Them Work Safely.
Your employees are already asking ChatGPT, Claude, Gemini, Copilot, and coding assistants to draft answers, summarize documents, write scripts, and speed up daily work. The gap appears when those useful experiments need approved data, repeatable prompts, project folders, tool access, audit trails, and a human approval path before anything touches a customer or production system.
ITECS builds that missing operating layer. We create custom AI agents and guided workflows that can retrieve the right context, call approved tools, connect to external systems, request human approval, and log what happened. Sometimes that means a Claude project folder or Codex workflow. Sometimes it means a full agentic RAG system with integrations, guardrails, and AI DevOps.
Illustrative Planning Scenario
This modeled example shows how an engagement could be scoped; it is not presented as a reported client result.
A 55-person property management company in Dallas: could have front-desk staff repeatedly searching separate systems for lease, maintenance, and payment information while useful prompt experiments remain disconnected from approved records and review rules.
Modeled outcome: A governed agent could retrieve approved material through scoped connections, draft an answer, and escalate policy or account-specific decisions with source context for staff review. Resolution targets would be validated during discovery.
Anatomy of an Agent Run
Watch one request become a governed action
Scroll through a single request as a custom agent handles it. Each move is a control ITECS designs, tests, and documents before an agent touches a live system.
Illustrative run: the moves are the design pattern. Your workflow, systems, and approval rules are defined during discovery.
The request arrives with an identity
A staff member asks the agent to prepare a customer follow-up. The agent confirms who is asking and which folders, workspaces, and systems that person may use.
It retrieves approved context first
Agentic RAG pulls only approved documents, tickets, and records. Answers carry source citations, so staff can check the evidence before anything moves.
It calls only the tools it is allowed to use
The agent plans the next step and calls scoped tools: a CRM lookup, a helpdesk search, a file store, an API, or an approved CLI workflow.
Guardrails check the plan
Permissions, data boundaries, confidence thresholds, and DLP rules are checked before the agent acts. Out-of-policy or low-confidence steps route to a person instead of guessing.
A person approves the action
Sensitive actions wait in an approval queue with the draft, the sources, and the proposed change. Staff approve, edit, or reject. Nothing is sent, changed, or executed until they do.
It acts, logs, and stays monitored
Only approved actions run. Each step lands in an audit log, and monitoring tracks quality, cost, and drift after launch.
Capabilities
Custom AI Agent Capabilities
Claude, ChatGPT, Gemini, Copilot, Codex, CLI, and open-source agent workflows
Agentic RAG systems that retrieve approved context before answering or acting
Secure connections to CRMs, helpdesks, file stores, databases, codebases, and APIs
Human-in-the-loop approvals for sensitive messages, records, transactions, and production changes
Project folders, prompt systems, tool schemas, evaluation sets, and operating documentation
Compliance-eligible deployment options scoped to the client's obligations and vendor terms
How to build custom AI agents for your business
- 1
Map the workflow, tools, data, and approval points
We identify which tasks should be assisted, automated, or left human-owned. Then we map the folders, apps, APIs, databases, prompts, approvals, and security controls the agent needs.
- 2
Build the agent with the right AI tools
We use the right mix of AI applications, APIs, RAG pipelines, CLI tooling, prompt systems, tool calls, and workflow automation. That can include Claude projects, Codex workflows, ChatGPT, Gemini, Microsoft Copilot, or custom-built agent services.
- 3
Test, deploy, train, and monitor
We validate accuracy, permissions, guardrails, logs, costs, and escalation paths before launch. Then we train your staff and monitor the agent so it keeps improving safely.
Custom AI Agent Pipeline
From Workflow Design to a Governed Agent in Production
Map Workflows
Tasks, folders, tools, data
We identify which tasks to assist, automate, or leave human-owned, and the approvals the agent needs.
Build Agent
Prompts, tools, RAG, CLI logic
The agent is built with the right mix of AI applications, APIs, RAG pipelines, CLI tooling, and tool calls.
Test & Tune
Approvals, guardrails, evals
Accuracy, permissions, guardrails, logs, costs, and escalation paths are validated before launch.
Deploy
Claude, Codex, Slack, web
It ships where your team already works, and staff are trained on how to use and escalate it.
Monitor
Logs, costs, actions, quality
Monitoring keeps it improving safely after launch.
Agent Designs
Agent designs you can inspect before you commission one
The first four are defined ITECS service and solution designs, not reported client case-study outcomes. The last two are agents ITECS built and runs for its own operations.
Manufacturing finance
PPV Agent
Watches purchase price variance and commodity cost exposure for finance and procurement, with human approval before any financial action.
Explore the PPV Agent designAsset-based lending
Field Exam Analyzer
Turns borrower agings, inventory, GL, and bank data into a first-draft working-capital exam that the examiner reviews and signs.
Explore the Field Exam Analyzer designKnowledge and IT operations
ITECS OpsMemory
Turns approved company knowledge and vendor guidance into cited answers and maintained SOPs, with human review for consequential changes.
Explore ITECS OpsMemoryIT change readiness
ITECS Change Assurance
A read-only review agent that stress-tests infrastructure change plans and produces an auditable readiness verdict. It never makes the change.
Explore ITECS Change AssuranceBuilt for ITECS
On-call scheduling agent
A Microsoft Copilot agent in Teams that checks on-call coverage, creates events, and automates shift-swap approvals, tied into the ITECS PSA.
Read how ITECS built the on-call agentBuilt for ITECS
Shared Codex and Cowork agents
ITECS distributes Codex and Claude Cowork agent files through self-hosted Seafile while governing model, tool, and client-data flows separately.
Read how ITECS shares agents across the teamPlatforms We Build On and Connect To
- Claude
- Anthropic API
- ChatGPT
- OpenAI API
- Codex
- Gemini
- Microsoft Copilot
- Azure OpenAI
- GitHub
- Slack
- Microsoft Teams
- HubSpot
- Salesforce
- HaloPSA
Security
Enterprise-Grade Security for Business Data
Custom AI agents can touch customer data, internal policies, source code, credentials, and proprietary business systems. ITECS AI is backed by ITECS — a Dallas cybersecurity MSP since 2002 — and credential, access, and compliance posture is reviewed alongside our cybersecurity advisory team when an agent handles regulated or high-value data.
Vendor-neutral architecture — we build with the right mix of ChatGPT, Claude, Gemini, Copilot, Codex, APIs, open-source models, and business platforms for your use case
Human-in-the-loop controls — sensitive actions can require approval before an agent sends a message, updates a record, triggers a workflow, or touches production
Data boundaries — agents retrieve from approved folders, systems, and knowledge bases with scoped access, confidence thresholds, and audit logging
Scoped protection — prompts, retrieval data, tool calls, logs, and credentials use controls and retention policies selected for the deployment architecture
Agent Control Stack
Seven controls between an agent and your production systems
Each layer answers one question an agent must clear before it acts. The linked ITECS guides go deeper on every control.
Inventory every agent and dependency
An AI bill of materials inventories what each agent depends on before it moves to production.
Trust only pinned, scoped tools
Governed registries, pinned versions, scoped access, and drift detection keep poisoned tools, plugins, skills, and MCP servers out.
Authorize sequences, not single actions
Authorization judges the chain of tool calls an agent makes, not each action alone.
Bound every delegation and handoff
When work moves between agents and people, bounded task contracts, least-privilege access, and evidence-bearing handoffs keep an accountable owner in place.
Inject secrets at runtime
Scoped runtime injection, for example through the 1Password CLI, reduces how often credentials sit in plain text around agent workflows.
Score tool use before production
Trace-based evaluation scores outcomes, tool use, safety, reliability, latency, cost, and human review before an agent handles live work.
Stop, contain, and recover
An emergency stop, an incident playbook for containment and evidence, and backups an agent cannot delete define what happens when something goes wrong.
AI kill switch plan AI agent incident playbook AI agent recovery plan
Pricing
How Much Does a Custom AI Agent Cost?
Useful one-off prompts need a governed system before they access business data, call tools, or change records. Here is the design comparison ITECS validates when an experiment moves toward production.
Planning comparison only. Scope, timing, and outcomes vary by data, workflow, approvals, adoption, and implementation conditions.
DIY Prompt or No-Code Agent
ITECS Custom AI Agent
Workflow design
Prompt saved in one person's account
Documented process, roles, approvals, and success criteria
Tool access
Manual copy and paste
Scoped API, CLI, folder, and app integrations
RAG and context
Generic model memory
Approved knowledge, citations, and retrieval controls
Human escalation
Generic fallback message
Approval queues and full context for staff
Agent actions
Drafts text only
Can perform approved tasks and update systems
Ongoing operation
Degrades without updates
Monitored, tested, versioned, and tuned
Outcome hypothesis to validate
The right custom agent should remove handoffs, reduce errors, and give staff leverage without removing human judgment from sensitive business decisions.
This is not a reported client result or guarantee. Baseline the current workflow and measure accepted work before relying on it.
- Agent Discovery & Technical Specification: $4,500–$7,500, required first and credited toward the build
- Proof of Concept / Prototype: $8,000–$18,000 for bounded feasibility work without a production SLA
- Single-Workflow Production Agent: $18,000–$35,000 for one governed workflow in production
- 1 production agent: $2,500/mo after launch; production operations are separate from the build fee
- Claude projects, Codex workflows, CLI automations, RAG agents, and external-system integrations are all in scope
- Regulated environments are custom-scoped on compliance-eligible platforms
Agent Operations
After launch, a production agent is operated, not abandoned
Ongoing operations for production agents after launch, separate from Managed Intelligence Services adoption tiers. ITECS runs the checks that keep a production agent accurate as data, tools, and usage change.
Monitoring and evaluations
Agent runs, tool calls, costs, and answer quality are monitored and re-scored against the evaluation set.
Quality and drift checks
Changes in data, prompts, or model behavior are caught before they erode results.
Data refresh and exception triage
Retrieval sources stay current, and exceptions route to the right person with context.
Production tuning based on usage
Prompts, tools, and thresholds are tuned from real usage, with changes versioned and tested.
Agent Operations pricing
- $2,500/mo 1 production agent
- $4,500/mo 2 production agents
- $6,500/mo 3 production agents
- Custom quote 4+ agents, aggressive data-refresh cadence, or tight review SLAs
Production operations are billed separately from the build fee.
AI DevOps for production agents Managed Intelligence services Published AI pricing
Engagement Targets
Illustrative targets to baseline and validate during discovery — not reported client results or performance guarantees.
40%
Fewer Manual Handoffs
85%
Target Task Resolution Rate
3sec
Avg. Retrieval Time
Related work · An anonymized client account
Guided AI Builds: Learn by Building a Planning Agent
How ITECS and a client employee used four guided sessions to build a planning agent, turning business knowledge into reusable reports and review controls.
See what we built alongside the employeeFAQ
Custom AI Agents FAQ
How much does a custom AI agent cost for a growing organization?
Every production build starts with Agent Discovery & Technical Specification at $4,500–$7,500, credited toward the build. Published phases then range from $8,000–$18,000 for a bounded prototype to $55,000–$120,000 for a multi-agent system or major process redesign. Final scope is confirmed before work begins, and production operations are separate.
Can you build workflows for Claude, Codex, ChatGPT, Gemini, and other AI tools?
Yes. ITECS is not tied to one AI vendor. We build custom workflows for AI applications, APIs, CLIs, project folders, and agent frameworks, including Claude, Codex, ChatGPT, Gemini, Microsoft Copilot, Azure OpenAI, and other business AI platforms when they fit the use case.
What is an agentic RAG agent?
An agentic RAG agent retrieves approved context from your documents, databases, tickets, or systems before it answers or acts. Unlike a basic chatbot, it can use tools, follow multi-step instructions, request human approval, cite sources, and update external systems when permitted.
How do I secure my business data when using AI agents?
The biggest risk is connecting AI to sensitive data without access controls, logging, or approved workflows. ITECS scopes private or enterprise configurations, permissions, DLP policies, credential isolation, audit logs, and training to the selected vendors' contractual data terms and the client's requirements.
Can a custom agent perform tasks, not just answer questions?
Yes. Agents can draft messages, summarize records, create tickets, update CRM fields, prepare reports, run approved CLI workflows, trigger automations, or route work to the right person. Sensitive actions can require human approval before anything is sent, changed, or executed.
Can the agent connect to our CRM, helpdesk, files, and internal systems?
Yes. We build integrations with HubSpot, Salesforce, HaloPSA, ConnectWise, Hudu, ServiceNow, Microsoft 365, Google Workspace, GitHub, databases, file stores, and custom APIs. The agent pulls only the data it is allowed to use.
Ready to see where AI moves your business forward?
1Send intake
2Scope the need
3Choose the next step