# Lua, in full > Lua is the workspace for AI agent teams. You see every step they take. ## Definitions - An agent does one job, such as reviewing every pull request or researching each new lead. - An agent team is a group of agents with different roles working on one function, such as sales or engineering, and handing work to each other. - An agent fleet is an agent team given a number to hit, such as issues fixed within a day or qualified meetings a month, with a supervisor agent that assigns work, checks it against the number and asks a person when a decision matters. - Lua Workspaces is the shared app where people and agent teams work: chats, the team, each agent's own app and an inbox for decisions. ## How it works 1. Create an agent from a short form: a name, what it does, and one or two tools suggested for your role. 2. Grow it into a team: add roles as the work grows, with a supervisor agent keeping the team on its number. 3. Each agent gets its own app that shows its work, live, for everyone you share it with. 4. Your people join in: comment, approve and set the rules in the same place the work happens. ## Control - Anything a customer sees, money spent or records changed waits for a person by default. - Governance rules block or pause named tool calls whatever the model decides, and every incoming message is scanned for prompt injection. - Several people can be approvers; the first decision wins and the others see who decided. - Roles: viewer, operator, builder, admin, owner and billing admin. - Every step and tool call is logged and can be sent to your own HTTPS endpoint, OpenTelemetry, Datadog, Better Stack, Splunk, Grafana Loki, Axiom, New Relic, Sumo Logic or object storage. - Each agent team has its own pool of actions. ## Agent teams ### Commercial (https://heylua.ai/agent-teams/commercial/) Goal: Qualified meetings a month. Agents that research accounts, find the reason to call now and draft every message for your approval, around one number: qualified meetings. Roles: Sales Director (Supervisor. Writes the weekly plan and keeps the team on the number.); Account Strategist (Researches each account and finds the reason to call now.); Outreach Writer (Drafts every first message and follow-up for a person to approve.); Revenue Manager (Keeps the pipeline current and reports the week.) Example: A new account enters the week. Account Strategist: Reads Wrenfield Health’s site, filings and its COO’s last three posts. 14 sources in 41 seconds. Account Strategist: Finds the reason to call now: a new head of operations started in September. Outreach Writer: Drafts a 118-word first message and checks the rules: no pricing promises, no attachments. Sales Director: Asks Dana to approve, because first messages to new accounts need a person. Dana changes one line and approves at 09:14. Revenue Manager logs the send in the CRM and sets a follow-up for Thursday. Wrenfield Health replies on Wednesday and a meeting is booked for Thursday at 14:00. Q: Does it send emails by itself? A: No. Sending is off until a person switches it on, and every message is drafted for your approval first. Q: Which CRMs does it work with? A: HubSpot, Salesforce, Pipedrive, Zoho CRM and the other CRMs in the catalogue, and any tool with an MCP server. Q: How does it pick accounts? A: It looks for accounts showing buying signals, ranks them and proposes the week’s list for you to approve. ### Growth (https://heylua.ai/agent-teams/growth/) Goal: Demos booked a month. A growth team of agents that researches, writes, runs ads and reports, around one number your marketing lead owns. Roles: Growth Lead (Supervisor. Sets the week’s bets and checks each against the number.); Researcher (Watches the market and finds the angle.); Writer (Drafts posts, pages and emails for approval.); Ads Manager (Runs paid tests within a budget you set.) Example: Three competitors cut their prices. Researcher: Reads 12 competitor sites at 07:00 and finds three price cuts since Friday. Writer: Drafts three versions of the pricing page, each with the enterprise line kept. Growth Lead: Picks the version that holds cost per demo under €70 and asks Maya to publish. Ads Manager: Pauses two ads running over €90 a demo, inside its €500 cap. Maya comments “keep the enterprise line”, Writer revises, Maya approves. The page is live at 11:02. Cost per demo this month: €62, down from €90. Q: Does it publish by itself? A: No. Posts, pages and emails wait for a person, and ad spend stays inside the cap you set. Q: Can it work with our brand voice? A: Yes. Give it your guide and examples, and comment on drafts to teach it. ### Customer Engagement (https://heylua.ai/agent-teams/customer-engagement/) Goal: Questions answered same day. Agents that answer customers on the channels they already use, and hand the hard ones to a person with the context ready. Roles: Service Lead (Supervisor. Watches queues and escalations.); Front Desk (Answers on WhatsApp, chat and email.); Order Desk (Checks orders and changes them within rules.); Escalations (Prepares the hard cases for a person.) Example: A WhatsApp message at 19:40. Front Desk: Reads “my order 4471 hasn’t arrived”, finds the order in the shop system and sees it is two days late. Order Desk: Rebooks delivery for tomorrow, inside the rule that allows free re-delivery under €50, and tells the customer. Escalations: Prepares a separate refund case over €500 for Lena, with the order history and the customer’s messages attached. Service Lead: Reports the day: 412 answered, 96% the same day, 5 escalated. Lena approves the €640 refund from her phone. The customer has a delivery date in four minutes and the refund in the morning. Q: Which channels? A: WhatsApp, Messenger, Instagram, web chat, email, SMS, voice and Slack, among others. Q: Who handles refunds? A: Refunds over your limit wait for a person. ### People (https://heylua.ai/agent-teams/people/) Goal: New starters ready on day one. Agents that run hiring admin and onboarding, so every new starter has their access, kit and first week ready. Roles: People Lead (Supervisor. Tracks every hire from offer to day 30.); Recruiting Coordinator (Schedules interviews and chases feedback.); Onboarding Coach (Runs the first week plan.); Policy Desk (Answers policy questions with sources.) Example: An offer is accepted on Monday. People Lead: Opens the starter plan for a product designer starting on the 3rd. Onboarding Coach: Orders the laptop, asks IT in Slack for the accounts and books the first-week sessions. Recruiting Coordinator: Books three of the four interviews for the account executive role and chases the fourth interviewer’s feedback. Policy Desk: Answers “how does parental leave work here?” with the policy and the page it came from. The offer letter for the next hire waits for Jo. Jo reads it and approves. Nine starters this month, all with kit and access on day one. Q: Does it make hiring decisions? A: No. It handles the admin. People make every decision. ### Back Office (https://heylua.ai/agent-teams/back-office/) Goal: Invoices matched within a day. Agents that read invoices, match them, chase payments and close the books faster, with every entry traced back to its source. Roles: Finance Lead (Supervisor. Watches the close and the exceptions.); Payables Clerk (Reads invoices and matches them to orders.); Receivables Clerk (Chases late payments politely.); Reconciler (Reconciles accounts and flags gaps.) Example: An invoice arrives as a photo. Payables Clerk: Reads invoice 2291 from a photo of a paper document, pulls out the supplier, amount and line items, and applies the right tax for that kind of cost. Payables Clerk: Matches it to purchase order 1184 and marks it ready for the team to check. Reconciler: Spots that the same supplier billed twice in March and raises it for Omar. Receivables Clerk: Sends customer 118 its second reminder, from the template Omar approved. The €8,200 payment waits for Omar. He checks the match and approves. 312 invoices matched this week, 98% within a day. The close takes four days instead of nine. Q: Can it pay suppliers? A: Payments always wait for a person. Q: Does it work with invoices that are photos? A: Yes. It reads the image, pulls out the fields and applies the right tax treatment, then a person checks it. ### Retail (https://heylua.ai/agent-teams/retail/) Goal: Stores kept in stock. Agents that watch stock across stores and distribution centres, flag the stores at risk and draft the orders to fix them. Roles: Supply Lead (Supervisor. Watches stock health across the network.); Store Analyst (Scores each store’s stock health.); Order Planner (Drafts replenishment orders.); Supplier Desk (Chases late suppliers.) Example: Monday at 06:00. Store Analyst: Scores every store’s stock health before the stores open. 52 are at risk, down from 71 last week. Order Planner: Drafts 12 replenishment orders for Centre North, sized to each store’s sales rate. Supplier Desk: Chases the supplier whose delivery to Centre North is two days late. Supply Lead: Asks Karim to approve the 12 orders before they go to suppliers. Karim approves 11 and holds one for a store that is closing. Orders go to suppliers at 08:40. Availability is 94%, against a goal of 96%. Q: Does it place orders itself? A: Orders are drafted for a person to approve. ### Property Management (https://heylua.ai/agent-teams/real-estate/) Goal: Enquiries answered in minutes. Agents that answer enquiries, book viewings and keep owners updated, across every listing. Roles: Lettings Lead (Supervisor. Keeps every listing moving.); Enquiry Desk (Answers and qualifies enquiries.); Viewings Coordinator (Books and confirms viewings.); Owner Updates (Writes the weekly owner report.) Example: Twelve enquiries overnight. Enquiry Desk: Answers each of the 12 enquiries on House, Elm Road with the details, the next viewing slots and two qualifying questions. Viewings Coordinator: Books three viewings on Saturday for Flat 3B and confirms each one by WhatsApp. Owner Updates: Writes the weekly report for the owner of 3B: enquiries, viewings and the offer that came in. Lettings Lead: Takes the offer on 3B to Rita, because offers are decided by a person. Rita accepts the offer at 10:20. 138 enquiries this week, answered in a median of four minutes. Q: Can it negotiate? A: No. Offers go to a person. ### Hospitality (https://heylua.ai/agent-teams/hospitality/) Goal: Guest messages answered in minutes. Agents that answer guests, handle changes and keep the front of house informed, day and night. Roles: Guest Lead (Supervisor. Watches the inbox and escalations.); Guest Desk (Answers guests on WhatsApp and email.); Reservations (Changes bookings within rules.); Reviews (Drafts replies to reviews.) Example: A guest writes at 23:10. Guest Desk: Reads a message in Portuguese asking for a late check-in, confirms it in Portuguese and tells the night team. Reservations: Moves a booking to a quieter room, inside the rule that allows changes at the same rate. Reviews: Drafts a reply to a three-star review that mentions a cold breakfast, for Nadia to approve. Guest Lead: Takes a request for a free upgrade to Nadia, because upgrades are her call. Nadia approves the upgrade to room 214 in the morning. 226 messages today, 98% answered, three for Nadia. Q: Which languages? A: Guests are answered in their own language. ### Engineering Fleet (https://heylua.ai/engineering-fleet/) Goal: Serious issues fixed within a day. Specialist agents assess, prepare and review engineering work. A supervisor agent watches delivery health and brings exceptions to your engineers. Roles: Engineering Manager (Supervisor. Tracks each agent’s numbers, routes exceptions, can roll back.); Code Reviewer (Reviews every pull request within minutes.); Security Engineer (Patches vulnerable packages and checks endpoints.); Reliability Engineer (Watches errors and latency, fixes the most frequent.) Example: A 5xx alert at 03:12. Reliability Engineer: Picks up the alert on chat-ingress (NW-103), reproduces it in staging and finds a null body on retries. Reliability Engineer: Opens a fix as a pull request with a test for the retry path. CI passes at 03:41. Code Reviewer: Reviews the pull request within minutes and leaves two notes, one on naming and one on a missing log line. Engineering Manager: Routes NW-142, an upgrade of five vulnerable packages, to Tom, because dependency upgrades need an engineer. Tom reads the diff over coffee and approves the merge. Nine serious issues open, down 14 in seven days. 212 closed this month. Q: Does it merge code by itself? A: No. Risky changes pause for a person, and an engineer approves what merges. Q: What does it need access to? A: Your issue tracker and code host, with the permissions you choose. Q: How do I get access? A: Apply on this page. We are opening it to a small number of teams. ### Lua Invest (https://heylua.ai/agent-teams/lua-invest/) Goal: Deals screened every week. An agent team for investment firms: screening deal flow, preparing diligence and drafting the investment memo for partners to decide. Roles: Investment Director (Supervisor. Keeps the pipeline moving and the memos sharp.); Deal Scout (Screens inbound and sourced deals against your thesis.); Diligence Analyst (Builds the research pack for each deal.); Portfolio Monitor (Watches portfolio companies and flags changes.) Example: Eighteen decks on Monday. Deal Scout: Scores 18 new decks against the thesis. Caldbeck Robotics scores 82: strong team, right stage, one open question on margins. Diligence Analyst: Builds the Lowick Payments pack from the data room and public filings, and drafts the memo with every source linked. Portfolio Monitor: Flags a new CFO at Thornby Bio and attaches the announcement. Investment Director: Asks Elena to review the Lowick memo before it goes to the investment committee. Elena reads the memo, adds a question on churn and approves. The memo is shared with the committee. 142 deals screened this month, 18 above the bar. Q: Does it make investment decisions? A: No. It screens, researches and drafts. Partners decide. Q: Can it work with our data room? A: Yes, with the access you grant it. ## Customers ### Turaco (https://heylua.ai/customers/turaco/) Same team, three to four times the customers: Turaco’s claims desk on WhatsApp. Turaco is a Nairobi-based insurer whose customers mostly use WhatsApp for their policies and claims. A Lua agent now meets those customers first. The same team of under 30 people serves three to four times as many of them. Quote: "An agent can respond to multiple customers. At a go." (Charles Opuba, who leads claims quality at Turaco). 3 to 4× the customers served by the same team of under 30; Under a minute first replies that used to take around 15 minutes; A third of the time to close a claim; 24 hours open in Kenya and Uganda, holidays included. ## Answers ### How can I see what our AI agents are doing and step in when they get something wrong? (https://heylua.ai/answers/see-what-ai-agents-are-doing/) Give each agent an app that shows every step it takes, in plain words, as it happens, and make anything that matters wait for a person. In Lua each agent has its own app, every run leaves its steps behind, and your team comments, approves or stops the work in the same place. How it works: Each step an agent takes is recorded as it happens, such as “Searched the CRM for the account” or “Drafted the first message”. The agent’s app shows those steps beside the item it is working on, and an activity feed shows the team’s work in order. A person can open any item, read the steps and comment. What a person keeps: Anything a customer sees, money spent or a record changed waits for a person by default. Rules can block or pause any tool call whatever the model decides. A comment on a draft changes the next draft, and a rule written in a comment applies from then on. What to watch for: An agent you cannot read is an agent you cannot correct. If a tool shows you only the final answer, ask where the steps are. If it shows you logs only a builder can read, ask how a head of sales would use them. Example: Dana opens the Pipeline desk at 09:00 and sees that the Outreach Writer drafted a first message to Wrenfield Health at 08:41, after the Account Strategist read 14 sources. She changes one line and approves. The step, her change and the send are all in the record. Q: Can I see what an agent did last week? A: Yes. Every run keeps its steps, and the log can also be sent to your own tools. Q: Can I stop an agent mid-run? A: Yes. A rule can pause any tool call, and a person can stop the work from the app. Last checked 27 October 2026. ### Can a small team get more done with AI agents instead of hiring? (https://heylua.ai/answers/ai-agents-instead-of-hiring/) Yes, for the work that repeats: reading, sorting, drafting, chasing and updating. An agent takes a whole job of that kind, shows its steps, and asks a person before anything important goes out. The people you have spend their time on the decisions. How it works: Start with one job you do every week and one tool it lives in. Create an agent with a name and what it does, connect the tool, and read its steps for a week. When the queue fills with different kinds of work, split it into roles and add a supervisor agent that keeps the team on one number. What a person keeps: People make the decisions: what the agent may do alone, what waits for approval, and which number the team owns. An agent team does not replace a person’s judgement; it removes the repetition around it. What to watch for: Agents that work in a chat thread leave the work in the thread. Look for an app your whole team can open, and for a record of steps anyone can read. Count the time saved only once the agent runs for a month. Example: Turaco’s claims desk, a team of under 30, serves three to four times as many customers on WhatsApp since a Lua agent started meeting customers first. First replies often arrive in under a minute. Q: Which jobs work first? A: Jobs with a clear input and a clear output: inbound leads, invoices, tickets, enquiries, reviews. Q: Does it need a developer? A: No. You create an agent from a short form. Developers can build the same agents in TypeScript. Last checked 27 October 2026. ### Should I hire another SDR or use AI agents for outbound sales? (https://heylua.ai/answers/ai-agents-or-another-sdr/) Use agents for the research and the drafting, and keep people for the conversations. A commercial agent team researches each account, finds the reason to call now and drafts every message for a person to approve. The person sends, talks and closes. How it works: An Account Strategist agent reads each account’s site, filings and recent posts and writes a brief. An Outreach Writer drafts the first message and follow-ups against your rules. A Sales Director agent keeps the team on one number, such as qualified meetings a month, and asks a person before a first message to a new account goes out. What a person keeps: Sending is off until a person switches it on. Every message is a draft until someone approves it. The CRM is updated by the Revenue Manager agent, so the pipeline is current without anyone typing. What to watch for: An agent that sends on its own will send something you would not. Keep the approval on first messages for as long as it takes to trust the drafts, and keep the person on every reply. Example: In the Commercial team’s demonstration, the week’s list of 24 accounts is researched by Monday morning, three drafts wait for Dana, and the meeting with Wrenfield Health is booked on Thursday. Q: Does it work with our CRM? A: HubSpot, Salesforce, Pipedrive, Zoho CRM and the other CRMs in the catalogue, and any tool with an MCP server. Q: How many meetings will it book? A: That depends on your list and your offer. The team owns the number and shows you what moved it each week. Last checked 27 October 2026. ### What is an agent fleet, and how is it different from an agent team? (https://heylua.ai/answers/what-is-an-agent-fleet/) An agent fleet is an agent team given a number to hit, such as issues fixed within a day or qualified meetings a month. The fleet has a supervisor agent that assigns the work, checks it against the number and asks a person when a decision matters. How it works: A single agent does one job. An agent team is a group of specialists around one function, each with a job title. A fleet adds the supervisor and the number. The supervisor splits the goal into jobs, the specialists do the steps in your tools, and the supervisor checks progress and changes the next step. What a person keeps: You decide where a person is needed: sending to a customer, spending money, changing a record or merging code. While a step waits for you, the fleet carries on with the rest of the work. What to watch for: A fleet without a number is a team. A fleet with three numbers is three fleets. Pick one number the team can move within a month. Example: Lua’s Engineering Fleet owns “serious issues fixed within a day”. A Reliability Engineer agent fixes the most frequent errors, a Code Reviewer reviews every pull request within minutes, and the Engineering Manager agent routes the merges that need an engineer to Tom. Q: Who coordinates the fleet? A: The supervisor agent. People decide at the points that need a person. Q: Can a fleet act without approval? A: Anything a customer sees waits for a person by default, and you choose what the fleet may do alone as trust grows. Last checked 27 October 2026. ### Why would my team use an AI agent workspace instead of just using ChatGPT or Claude? (https://heylua.ai/answers/agent-workspace-or-ai-assistant/) An assistant helps you do the work, and you do the steps. An agent workspace is where agents do the steps in your tools, show them in an app your whole team can open, and ask a person before anything important goes out. The work stays shared, recorded and repeatable. How it works: In a workspace, an agent has a job, tools and rules that stay the same every time. Its work shows up in an app, not in one person’s chat history. Several agents work as a team, hand work to each other and keep one number in view. People join in to approve, comment and set rules in the same place. What a person keeps: Your assistant stays useful for your own work. The workspace is for work the company owns: the leads, the invoices, the tickets, the claims. Every step is recorded, and a person approves what matters. What to watch for: If the only record of an agent’s work is a thread, nobody else can check it, and the agent forgets the rules you gave it last week. Look for apps, steps and rules that persist. Example: In Lua, the Commercial team’s Pipeline desk shows 24 accounts, who is working on each, and the three drafts waiting for Dana. Sam, Dana’s account executive, opens the same app and sees the same steps. Q: Can I still chat with an agent? A: Yes. Every app has the agent in a side pane, and each person has their own conversation with it. Q: Which models does it use? A: Models from OpenAI, Anthropic and Alibaba, chosen per agent, and you can switch at any time. Last checked 27 October 2026. ### How do AI agents triage bugs and prepare fixes overnight without engineers on call? (https://heylua.ai/answers/ai-agents-for-engineering-maintenance/) A fleet of specialist agents watches errors, reproduces them, opens a fix as a pull request with tests, and reviews it, while a supervisor agent routes anything that needs an engineer to the morning. Nothing merges without a person. How it works: A Reliability Engineer agent picks up the alert, reproduces the error in staging and opens a pull request with a test. A Code Reviewer agent reviews it within minutes. A Security Engineer agent upgrades vulnerable packages. The Engineering Manager agent tracks each one’s numbers and brings exceptions to an engineer. What a person keeps: Engineers approve what merges. The fleet needs your code host and issue tracker with the permissions you choose: read access first, then the right to open pull requests. Risky changes pause for a person. What to watch for: Coding assistants help an engineer write code. A maintenance fleet works the queue on its own and shows you the steps. Ask which one you are buying, and ask to see the record of a night’s work. Example: At 03:12 a 5xx alert lands on chat-ingress. By 03:41 the fix is a pull request with a passing build and two review notes. At 08:00 Tom has two things that need him, not twelve. Q: Does it merge code by itself? A: No. Risky changes pause for a person, and an engineer approves what merges. Q: Which tools does it need? A: Your code host (GitHub, GitLab, Bitbucket, Azure DevOps) and your issue tracker (Linear, Jira and others). Last checked 27 October 2026. ### How do I keep the CRM updated without asking sales reps to do it? (https://heylua.ai/answers/keep-the-crm-updated/) Give an agent the job. A Revenue Manager agent reads the emails, calls and meetings the team already has, updates the records, logs the activity and sets the follow-ups, and a person checks what it changed. How it works: The agent connects to the CRM and the inbox or calendar. After each email or meeting it updates the stage, the next step and the notes, and it drafts the follow-up. Every update is a step a person can read and undo. What a person keeps: Record changes can wait for a person at first, then run alone once the updates are right. The agent never sends a message without an approval, and the CRM keeps its own history of who changed what. What to watch for: Automation that writes to the CRM without showing its steps is how pipelines go wrong quietly. Read the steps for the first two weeks. Example: After the meeting with Greyline Freight on Thursday, the Revenue Manager agent moves the deal to “Proposal”, logs the notes and drafts the follow-up for Dana to approve. The pipeline is €410k and current. Q: Which CRMs? A: HubSpot, Salesforce, Pipedrive, Zoho CRM and the other CRMs in the catalogue. Q: Can it read our inbox? A: With the access you grant it, and only the mailboxes you choose. Last checked 27 October 2026. ### How do I stop deals going cold when my sales team forgets to follow up? (https://heylua.ai/answers/stop-deals-going-cold/) Let an agent own the follow-ups. It watches every open deal, drafts the next message on the day it is due, and brings it to the person who owns the deal to approve. Nothing is forgotten, and nothing is sent without a person. How it works: The agent reads the CRM and the inbox, works out which deals have gone quiet and why, and drafts a follow-up that refers to the last conversation. The person approves, edits or skips it. The agent logs the send and sets the next date. What a person keeps: The deal owner keeps every send. A founder running sales alone sees one list each morning: the follow-ups that need a decision, with the drafts ready. What to watch for: Sequences that fire on a timer send the same message to everyone. An agent that reads the deal writes a different message for each one, and shows you why. Example: On Monday the Pipeline desk shows six deals with no reply in ten days, each with a draft. Dana approves four, edits one and skips one that closed on Friday. Q: Does it need a RevOps person? A: No. The agent does the tracking. A person approves the messages. Q: Does it work for a founder-led team? A: Yes. One person can run it with one agent and grow it into a team later. Last checked 27 October 2026. ### Should we automate our team’s repetitive work with a workflow tool or with an AI agent platform? (https://heylua.ai/answers/workflow-tool-or-agent-platform/) Use a workflow tool when the steps are known and fixed. Use an agent when each piece of work needs reading and judgement, and you want a person to approve the result. Many teams use both: the workflow moves the data, the agent does the thinking. How it works: A workflow tool runs the same steps every time: when a form is filled, add a row and send an email. An agent reads the form, decides what kind of request it is, asks a question if something is missing, drafts the reply and shows a person the steps. Lua runs workflows too, with approval steps inside them. What a person keeps: With an agent, people set the rules and approve what matters. With a workflow, people fix the workflow when the input changes. Both leave a record; the agent’s record says why. What to watch for: A workflow that handles one case well breaks on the first unusual case. An agent that handles unusual cases needs a record of steps and an approval, or it will handle one of them badly without telling you. Example: Invoices that arrive as photos of paper from Uganda cannot be handled by a fixed workflow. A Lua agent reads each one, works out the tax for that kind of cost and gets it ready for the team to check. Q: Can an agent call a workflow? A: Yes. Lua agents run workflows, and workflows can include approval steps for a person. Q: Can I build the agent without code? A: Yes, from a short form. Developers can also build in TypeScript. Last checked 27 October 2026. ### Is it safe to give an AI agent access to our code repositories? (https://heylua.ai/answers/ai-agent-github-access-soc2/) It is safe when the agent’s permissions are set by a person, every step it takes is recorded, rules can block any action whatever the model decides, and an engineer approves every merge. Ask any vendor for those four things in writing. How it works: In Lua, a person sets what each agent may do with each tool. Governance rules block or pause named tool calls and scan every incoming message for prompt injection. Every step and tool call is logged, and the log can be sent to your own monitoring tools. Roles for viewers, operators, builders, admins and owners decide who can change any of it. What a person keeps: Engineers keep the merge. The fleet gets read access first, then the right to open pull requests. Nothing deploys without a person. What to watch for: Ask for the audit trail, not the certificate alone. A certificate says the vendor has controls; the trail shows you what the agent did in your repository last night. Example: The Security Engineer agent upgrades five vulnerable packages and marks the pull request merge ready. The Engineering Manager agent routes it to Tom, because dependency upgrades need an engineer. Tom approves, and the merge is in the log. Q: Where can I find compliance details? A: Ask us for the current report and data handling summary. Q: Can I send the logs to our own tools? A: Yes: your own HTTPS endpoint, OpenTelemetry, Datadog, Better Stack, Splunk, Grafana Loki, Axiom, New Relic, Sumo Logic or object storage. Last checked 27 October 2026. ### What should an AI pull request reviewer do, and what should a person keep? (https://heylua.ai/answers/ai-pull-request-review/) An AI reviewer should read every pull request within minutes, run the checks, and leave specific notes on correctness, tests and risk. A person keeps the merge, the design decisions and the final word on anything the reviewer flags as risky. How it works: The Code Reviewer agent reads the diff and the linked issue, runs the checks, and leaves notes a person would recognise: a missing test, an unhandled case, a naming slip. Risky changes are marked for an engineer. The review is a record of steps, so you can see what it read and what it ran. What a person keeps: Merging. Changes to how the system is designed. Anything the reviewer is unsure about. Over time, let the reviewer approve the small changes alone and read its record afterwards. What to watch for: A reviewer that comments on everything is ignored within a week. Look for specific notes and a clear line between “fine” and “needs a person”. Example: The Reliability Engineer agent opens a fix for NW-103 at 03:41. The Code Reviewer leaves two notes, one on naming and one on a missing log line, and marks the change safe. Tom merges in the morning. Q: Which code hosts? A: GitHub, GitLab, Bitbucket and Azure DevOps. Q: Does it replace code review by people? A: No. It does the first pass on every pull request. People decide what merges. Last checked 27 October 2026. ### Which AI agent platforms serve companies in the UK and Ireland? (https://heylua.ai/answers/ai-agent-platform-uk-ireland/) Lua is the workspace for AI agent teams, built by a team in London and Nairobi and backed by Norrsken22, Flourish Ventures, Phosphor Capital and Y Combinator. It works with the tools UK and Irish companies already use, and a person approves what matters. How it works: Create an agent for one job, grow it into a team, and give the team its own app your people can open. Lua runs the infrastructure, the channels (WhatsApp, Slack, Teams, email, SMS, voice and web chat) and the integrations, with the catalogue counted at 809 on 27 October 2026. What a person keeps: Your people set what each agent may do, approve what matters and can read every step. Logs can be sent to your own tools. What to watch for: Ask where the data is processed and who the subprocessors are. Lua publishes its privacy notice and subprocessor list on heylua.ai. Example: Lua’s own commercial, growth and engineering teams run on Lua, in one Workspace, with the people who own each number approving what goes out. Q: Where is Lua based? A: Lua Global Inc is a Delaware company. The team is in London and Nairobi, with customers in Africa, Europe and the US. Q: How do I get in? A: Join the waitlist for Workspaces, or book a demo to set up a team with us. Last checked 27 October 2026. ## Questions Q: What is Lua? A: Lua is one app where your people and their agent teams work together. Every step the agents take is visible, and your team can comment on it or step in. Q: How is this different from an AI assistant? A: An assistant helps you do the work, and you do the steps. Lua’s agent teams do the steps in your tools, in an app your team can see, and they ask a person before anything important goes out. Q: What is an agent team, and what is a fleet? A: An agent does one job. An agent team is a group of agents working on one function, such as sales or engineering. A fleet is a team given a number to hit, with a supervisor agent keeping it on track. Q: Do I need to code? A: No. You create an agent with a name, what it does and one or two tools suggested for your role. Developers can also build in TypeScript with the Lua CLI. Q: Which tools does it work with? A: Hundreds of tools, from CRMs, inboxes and calendars to finance and hiring tools. If a tool is missing, you can connect its MCP server. Q: When can I get in? A: Workspaces opens in waves, and we let people in by fit. Join the waitlist, or book a demo to set up a team with us. Q: What is Lua Workspaces? A: A shared app where people and AI agents work together: chats, the team, each agent’s own app and an inbox for decisions. Q: Who can see the agents’ work? A: Everyone in the Workspace you share it with. An app can be shared with people who cannot see the agent behind it. Q: Which models does it use? A: Models from OpenAI, Anthropic and Alibaba, chosen per agent, and you can switch at any time. Q: Can I bring my team? A: Yes. Once you are in, bring two or three colleagues. Q: What is the difference between an agent team and a fleet? A: An agent team is a group of specialists around one function. A fleet adds a supervisor that assigns the work, checks it against a number and asks a person when a decision matters. Q: Can a fleet act without approval? A: Anything a customer sees waits for a person by default, and you choose what the fleet may do alone as trust grows. Q: Who coordinates the fleet? A: The supervisor agent coordinates the specialists. People decide at the points that need a person. Q: How is the agent platform priced? A: By responses each month, with a free plan and paid plans from $25 a month. Q: How is Workspaces priced? A: Teams in early access are told the price before they are charged. Q: Can an agent team have its own budget? A: Yes. Each agent team can have its own pool of actions. Q: Can I control what each agent can do? A: Yes. Approvals are built in, governance rules can block or pause any tool call whatever the model decides, and a person sets what each agent may do with each tool. Q: Is every step recorded? A: Yes. Every step is logged and can be sent to your own monitoring tools. Q: Where can I find compliance details? A: Ask us for the current report and data handling summary. Q: Which languages? A: TypeScript, with the Lua CLI. People who do not code create an agent from a short form in Workspaces. Q: Do code agents show up in Workspaces? A: Yes. Agents you build in code work in the same Workspace, with the same apps and approvals. ## Company Lua Global Inc, a Delaware company. Founded by Lorcan O'Cathain (co-founder and CEO) and Stefan Kruger (co-founder and CTO). Seed round of $5.8M led by Norrsken22, announced 16 April 2026, with Flourish Ventures, Phosphor Capital and Y Combinator (YC F25). Website: https://heylua.ai. Docs: https://docs.heylua.ai. Blog: https://blog.heylua.ai.