Custom AI agents

Agents for the work your team keeps postponing.

Not a chatbot. Not a platform rollout. A bespoke agent that runs your workflow.

Between adding another senior engineer and deploying ArmyAI, there is a third lane: custom AI agents built around how your company actually operates. At Tunix, we design, build and deploy agents for software delivery, business operations, time-intensive back-office work, and accounting — owned by you, integrated with your tools, auditable end to end.

Scope your first agent

The missing lane

Régie adds people. ArmyAI runs delivery. Custom agents automate the rest.

You do not always need another hire or a full platform. Sometimes you need one reliable agent that closes tickets, reconciles invoices, drafts test plans, or pushes CRM updates — every day, the same way, without burning senior time. That is what we build.

The problem

Generic AI demos do not survive Monday morning.

The backlog nobody owns

Reconciliations, reporting packs, ticket triage, document checks — work that matters but never wins the sprint. It piles up until someone senior does it by hand.

Chat without memory or guardrails

A prompt in a browser tab is not an agent. No durable context, no tool access, no sandbox, no audit trail. Teams try it once, then go back to spreadsheets.

You need ownership, not another SaaS

Your workflows, your data residency rules, your ERP and Git repos. You want agents you can inspect, tune and extend — not a black box that sends everything to a third party.

What we build

Agents tuned to real jobs — not slide-deck fantasies.

Software development

PR review runners, test scaffolding, refactor assistants, release-note drafts, dependency audit bots — agents that speak your repo, your CI, your definition of done.

Business operations

CRM hygiene, pipeline updates, client onboarding checklists, contract extraction, SLA reminders — the operational glue that keeps revenue moving.

Time-intensive enterprise tasks

Document processing, compliance pre-checks, procurement follow-ups, multi-system data pulls — repetitive work that eats calendar blocks across departments.

Finance & accounting

Invoice matching, expense categorisation, month-end prep, inter-company reconciliations, audit trail assembly — with human sign-off where it legally matters.

How we build

Production-grade patterns — OpenClaw structure, Hermes memory.

We ship ArmyAI. We run delivery régie. Our custom agents borrow the same OSS patterns we battle-test in production — not toy wrappers around a chat API.

OpenClaw

Clear agent anatomy

Each agent gets an explicit ROLE, SOUL and TOOLS definition — readable Markdown, versioned in Git. Skills live in dedicated folders. Code-executing agents run in a deny-by-default sandbox. Every run leaves an append-only event log you can audit.

Hermes

Memory that compounds

Agents remember context across sessions — user preferences, project facts, enterprise rules — in human-readable memory files, not opaque vectors. When memory fills up, the agent consolidates explicitly. Your team can read, edit and revert what the agent learned.

OpenClaw and Hermes are open-source agent projects we study and adapt. Your agents are bespoke deliverables — source included, no vendor lock-in on the workflow itself.

How we engage

From workflow interview to agent in production.

Discovery & scoping

1 week

We map the workflow, the systems involved, the failure modes, and the human checkpoints. You get a scoped agent brief with success criteria — before a single line of agent code.

Design & prototype

2–3 weeks

ROLE/SOUL/TOOLS, tool wiring, memory layout, sandbox policy. You test the agent on real inputs with our team in the loop.

Integration & hardening

2–4 weeks

Connect to your ERP, Git, ticketing, email or internal APIs. Observability, error handling, runbooks, and security review for your environment.

Handover & run support

Optional

Documentation, training for your team, optional quarterly tuning under régie. You own the agent; we stay available when the workflow evolves.

Why Tunix

We build agents for teams who ship for real.

ArmyAI is our lab

We operate an on-prem agent platform for software delivery. Custom agents inherit orchestration, memory and sandbox patterns proven under NDA — not copied from a blog post.

Régie keeps us honest

We embed senior engineers in client teams every week. We know which automations actually stick — and which ones die in the pilot deck.

UK entity, B2B/B2G ready

One contract, one escalation path. NDA-friendly delivery, compliance-aware scoping, and agents designed for environments where audit trails matter.