AI that runs the work. People who stay in charge.
We build and run the AI behind your business. The team that scopes it still maintains it in year three.
~/queue/today
> nothing sends, posts, pays or deletes without a person
Running in production. A person still approves what goes out.
KURK
A health and longevity company. Six agents on one knowledge base that audits itself every day for stale notes, dead links and rules an agent stopped obeying.
My Performance Doctor
A membership medical practice in Melbourne. An agent fleet, a command centre built for the founder, and one knowledge base behind both.
Also in production
A university career-development centre. Daily student outreach across dozens of schools, drafted by an agent and sent after a person approves.
A car-dealership software company. Marketplace posting for dealers, one isolated browser per dealer, no shared IPs.
A golf-tour operator. Lead research that lands in a sheet the owner reviews before any outreach goes out.
Our own fleet. The same setup runs LFG Labs day to day.
Also built
Built for a US tax and accounting firm: bank-statement bookkeeping, financial statements, IRS forms and client comms, every output checked by a person before it left.
Watch the statements being generated →Built for a home-services company: inbound customer service over SMS, with address checks, booking and a safe handoff to a person.
- build on
- OpenClaw · Claude Code · Hermes · LangGraph · Claude · Obsidian · Markdown · MCP
- connect to
- Discord · Slack · SMS · Asana · Stripe · Zapier · Make · HousecallPro · TaxDome · Webflow · Google Ads · GA4 · Meta · LinkedIn
Five owners. No staffing layer.
Eight years working together in Karachi, each of us with equity, and no handoff to someone you have not met.
Abdul Sami
Lead architect · Founded Xord in 2018 and A51 Finance
- now
- embedded in client businesses, putting agents into production
- in his words
- “I think in systems. I can write code, and drive big technical decisions.”
Anas Bin Sohail
Backend and multi-tenant systems
Abdul Samad
Frontend
Mubashir Ali
AI and retrieval
Kaif Ahmed
Integrations
Context is always the bottleneck. Wiring a few OpenClaw or Claude agents is easy.
Every agent session is like a new hire, it starts from zero, and what it knows on day one is what you wrote down. So we write it down first, in plain files that OpenClaw, Claude Code, Hermes or LangGraph can all read.
~/your-business/knowledge-base
> every change leaves a trail, roll it back when you need to
The agent constitution.
What each agent does and does not do, written in plain Markdown in Obsidian, and the first deliverable in every build.
Scenarios.
Ten to twenty real situations per agent, run daily, so we catch a bad edit to the rules before it reaches a customer.
A versioned knowledge base.
Everything the agents know, plus the data sources they read for what is true right now, versioned, so what changed and when is on record and you can roll it back.
Context proposals.
An agent that hits something the rules did not cover drops a suggested edit as a pull request, and a person merges it or does not.
Approvals.
Drafts only by default: a person approves anything that sends, posts, pays or deletes, and you decide which actions can run unattended.
Your team, trained.
We train the people who own the process to manage the knowledge base themselves, so the agents keep being useful after we leave.
Start with the process you already run.
For founder-led businesses with a real ops team and too many tools, where hours go into work a well-briefed system could do first.
When not to
- You want a tool bought and installed, not a system run.
- Nobody on your side owns the process yet.
- You need it to make clinical, legal or financial decisions on its own. We build the draft and the approval, never the decision.
~/where-we-have-built
> interview the people doing the work, not the owner alone
Buy one step at a time.
Every engagement starts with the audit, and the rest is what it usually leads to.
- Audit
Find where the hours go.
We interview the people who do the work, owner included, and map the workflows end to end in one to two weeks.
- Build
Build in milestones.
Each milestone has a capped price and a written acceptance test, and you pay when you accept it, never a deposit.
- Run
Keep it running.
A flat monthly fee covers hosting on your accounts, monitoring, model costs at cost and knowledge-base upkeep, and it does not move when your volume does.
- Handover
Take it in-house.
Everything already runs on your hardware and your accounts, so whenever you want it you keep the code, the knowledge base and the runbooks.
We do not commit to a number we cannot know before the build. Where a result depends on your data, the first week is a paid calibration and the number comes from that.
AI workflow audit.
For founders who cannot say where the hours go. We interview the people doing the work and hand back the most urgent fix on page one.
~/audit/report.md
> fix the worst thing first
Command centre.
For owners running the business from six tabs. We build one calm screen with pipeline against target, the tasks that need a person today, who is due for renewal and what the agents did overnight.
Knowledge base and agent fleet.
For operations that need several specialists behind one contact. We write the knowledge base first, then put one chief-of-staff agent in front of the specialists that read from it.
~/fleet
Reliability layer.
For teams whose agents already run and break when a token expires, a provider goes down or someone edits a prompt. If the host dies we move you without switching frameworks, and one client has already lost an entire host and come through it intact.
~/fleet/health
This was an exceptional engagement, helping me set up an Open Claw system that is tailored to my needs and security. Very high quality of work with a clear scope and results. I will definitely use him again for future projects. Thank you for your help.
Excellent experience overall. Very professional, easy to work with, and delivered great results. I truly appreciated his commitment and responsiveness. Highly recommended.
Five notes on how we run agents in production.
A knowledge base that audits itself
Every morning a set of scripts reads the whole knowledge base and the last day of agent output, tests each agent against its own rules, and posts five lines to the team channel.
Agents never hold your API keys
When an agent needs data from your CRM, accounting or inbox, we put a small service between them that holds the credential and hands over only the fields you approved.
Drafts only by default
Every agent we ship does the grind on its own, and none of them can send, post, pay or publish until a person says so.
Rehearse a year of agent behaviour in 22 minutes
Before a change reaches a production fleet we run it through an isolated copy of the live setup, against the real models, and read what the agents say.
Spend tokens only on judgement
If a script can decide a step with certainty the model should never see it, and for the steps that are left the model should see the smallest slice of evidence that settles them.
- Platform choice.
- The knowledge base is plain Markdown and works with OpenClaw, Claude Code, Hermes or LangGraph, and we have moved clients between them without rewriting it.
- Ownership.
- You own it, since it runs on your servers and your accounts under your model subscriptions, and at handover you keep the code, the knowledge base and the runbooks.
- Cost.
- The audit is a fixed price, builds are capped per milestone and paid on acceptance, and the run is a flat monthly fee that does not move with your volume, all quoted after the call because the number depends on what we find.
- Timeline.
- The audit takes one to two weeks, and a first build usually runs four to eight weeks in milestones with something working by the end of the first.
- Your data.
- Each agent sees only the fields its task needs, through a firewall, a test that tries to make it leak personal data fails the build if it manages to, and sensitive data can stay on a model running on your own hardware if you need it.
- Agents you already have.
- Taking them over is the reliability layer, and we audit what is there first and tell you what to keep.
- After we leave.
- Your team is trained to manage the knowledge base before we go, so the agents keep getting better because the feedback loop is theirs now.
Start with the audit.
Thirty minutes to find out whether it is worth doing. If it is not, we will say so.
Or write to sami@lfglabs.ai