> ## Documentation Index > Fetch the complete documentation index at: https://docs.apollospace.ai/llms.txt > Use this file to discover all available pages before exploring further. # Construction & engineering > Builders, contractors, and engineering firms running many sites at once: chase subcontractors and suppliers, keep drawings and documents organized, turn an inbound RFI into a tracked task, and flag the milestone at risk — with payments, contracts, and any engineering judgment always staying with a qualified professional. It's 7am and three sites are running at once. The structural sub has gone quiet since Friday and nobody knows if today's pour goes ahead. A client email asking for a revised plumbing drawing has been sitting in the inbox for two days. The window supplier promised delivery yesterday and never confirmed. And the latest version of the architectural set is scattered across three different WhatsApp threads, with no one sure which one is good. None of it is complicated — it's just a lot, across several fronts, with too few people to watch every sub, document, and deadline at the same time. **What leaves your plate:** chasing each subcontractor and supplier for status and dates, keeping drawings and documents organized in the right folders, turning a loose RFI into a tracked task with an owner, and flagging a milestone or delivery before it slips all start happening on their own — with context, within budget, and with payments, contracts, and every technical decision always waiting on a human. ## How it works, in one line In construction, the bottleneck is rarely the building — it's **keeping subs, documents, and deadlines under control across several sites at once, without letting a delay surprise the schedule.** The agent doesn't approve a payment or sign a contract, and never makes an engineering, safety, or structural call: it makes sure every front gets chased, every RFI becomes a task with an owner, every risk surfaces before it stops the site — and it hands you anything that needs judgment, money, or professional liability. ```mermaid theme={null} flowchart LR A[Trigger
new RFI, deadline nearing,
sub silent, end of day] --> B[Agent pulls context
from Company Brain + CRM] B --> C[Decides + uses tools
CRM, Boards, Outbound,
Routines, Docs, Folders] C --> D{Sensitive?
payment, contract,
engineering judgment} D -->|No| E[Chases / logs /
flags / files] D -->|Yes| F[Proposes + calls a human
with full context] E --> G[Becomes memory
the next front already knows] F --> G ``` ## Concrete flows ### 1. Chase subcontractors and suppliers — status and dates **Trigger:** an agreed front hasn't checked in by its promised date, or a supplier's delivery date is approaching without confirmation. The agent pulls the sub, the front, the site, and the agreed date from the [CRM](/en/features/crm), checks the [Company Brain](/en/features/brain) for that supplier's history — who tends to run late, the real lead time, the right contact — and assembles a clean chase: "confirming the 3rd-floor pour at the Aurora site, scheduled for the 28th — still on track?". The send goes out by [email or WhatsApp](/en/features/outbound) with [Marcus](/en/agents/marcus) handling it, and the option to **require your approval before each send** while you don't yet trust the tone. The reply comes back as an activity on the card and the front's stage updates — confirmed, late, rescheduled. The result: you don't find out the structure stopped when you arrive on site. You find out when the agent chases, **days ahead**, and it only calls you in if the sub goes silent or gives you a bad date. What to ask and when to chase lives in the Brain, not a fixed script. A front that blocks the critical path — foundation, structure — is chased earlier and more firmly than a finish with float. The agent pulls that cadence from the history and the schedule you've already ingested. ### 2. Keep drawings and documents organized in the right folders **Trigger:** a document arrives — a revised drawing, a spec, a permit, a site diary, a material invoice — by email, [WhatsApp](/en/integrations/whatsapp), or [chat](/en/features/chats). The agent reads the file, works out what it is, and files it in the right [folder](/en/features/pastas) — by site, by discipline (architecture, structure, plumbing, electrical), by revision — preserving the original and noting the version. In the same move, it ingests the content into the [Company Brain](/en/features/brain), so the next question ("what's the latest electrical drawing for the Aurora site?") has an answer right away, with the correct document — not the one from three revisions ago. ```mermaid theme={null} flowchart TD A[Document arrives loose
drawing / spec / invoice] --> B[Agent identifies
site, discipline, revision] B --> C[Files in the right folder
+ ingests into the Brain] C --> D{Supersedes
an earlier version?} D -->|No| E[Stays organized
and searchable] D -->|Yes| F[Marks as current revision
and keeps the history] E --> G[Becomes memory] F --> G ``` The win isn't "a bot that renames files." It's that **nobody builds off an old drawing** because the right version is always in the right folder — and the answer to a site question comes from the real document, not a guess in a WhatsApp group. ### 3. Turn an inbound RFI or request into a tracked task with an owner **Trigger:** a request for information arrives — a client RFI, a question from the site foreman, a scope-change request — by email, WhatsApp, or form. The agent reads the message, extracts what matters — which site, which discipline, what's being asked, the deadline, and who asked — and creates a [task](/en/features/boards) on the RFI board with the fields filled in and the original attachment along, **routing it to the right owner** for that discipline or front. If the RFI matches something already in the [Company Brain](/en/features/brain) — a drawing, a spec, a similar prior answer — the agent notes the reference on the card so whoever responds isn't starting from zero. **The RFI that becomes rework almost always sat untouched in an inbox.** Turning each request into a tracked task — with an owner and a deadline, the moment it arrives — is what stops a question from becoming a wrong wall, and lets the client see the matter is moving. ### 4. The routine that flags milestones and deliveries at risk **Trigger:** a [routine](/en/features/rotinas) the agent schedules for itself sweeps, every morning, the schedule milestones and open material deliveries across every site. The routine compares milestones and promised dates and separates what's **late or about to be late**. For each, the agent raises a [proactive notification](/en/features/rotinas) in your bell, prepares the chase to the sub or supplier (flow 1), and, if the delay threatens a client milestone, flags it on the site board. At end of day, [Athena](/en/agents/athena) — the org's chief of staff — pulls together what changed status, what was confirmed, what slipped, and what's due tomorrow into a short digest, in your bell and on the channel you use. With the [Composio integration](/en/integrations/composio), that summary can land in the site team's Slack or become a calendar event — without anyone compiling a spreadsheet by hand. **The delay that becomes a liquidated-damages claim was almost always visible in advance.** A routine never has a hectic Monday: it looks at the milestones every day, at the same time, and turns "nobody saw it" into an alert in your bell before the problem reaches the site. For new suppliers or materials, [Scout](/en/agents/scout) does the public web research — who supplies that material, price range, references — and returns a summary with its sources, without guessing. ## What stays human The agent **proposes**, never decides alone, when the matter is: * **Payment and progress billing** — paying a subcontractor or supplier, releasing a measurement, an advance, a refund, any outflow of money. * **Purchase approval** — committing the quantity and value of an order, choosing a supplier. * **Contract and signature** — any commitment that creates an obligation with a sub, supplier, or client. * **Any engineering, safety, or structural judgment** — sizing, approving a technical solution, signing off a structure, deciding on risk. This is **always** a qualified professional's call, never the agent's. * **A commercial proposal to a client** — the estimate or formal offer that goes out. * **Heavy external, legal, or regulatory communication**, governance decisions, and destructive actions. In those cases the agent prepares everything — the sub's history, the reference drawing, the draft chase — and **waits for your approval.** Every action stays in the auditable trail, with author and rationale, and autonomy grows like a ratchet per kind of action, not a leap. ## Where to start Upload to the [Company Brain](/en/features/brain) your active sites, your subcontractor and supplier contacts with lead times, and the drawings and project sets you consult every week. That's where the agent pulls the right chase, the right version, and the on-the-spot answer from. Connect a [channel](/en/integrations/whatsapp) and let the agent follow the open fronts and material deliveries and chase status. Start with human approval before each external message and loosen as it proves itself. Create a [routine](/en/features/rotinas) that sweeps milestones and deliveries every morning, and another that delivers the sites digest. One agent, one cadence, one channel — and autonomy grows slowly. ## Next steps The full office pattern that holds the build together — records, documents, chasing, and the summary that ties the ends. Following a front, deadline, and delivery end to end — of which a build is a many-fronts-at-once case. Drafting and editing specs, reports, and RFI responses — with the right version always at hand. Taking the record and the summary to where the team already works — Slack, Notion, calendar, and hundreds of apps.