Builds—answered.

We build you a custom AI agent for as little as $500. One agent, one job, built specifically for your business and your systems: screening your Indeed applicants, drafting responses to your inbound leads, auditing your vendor invoices, QA-ing designs against your own rules. It's a standalone tool, hard-coded to your setup, hosted and maintained by us. You use it as much as you want. There are no seats and no per-user charges.

This is a membership benefit. A comparable custom build from an AI dev shop runs $20K–$50K upfront. We can start at $500 because we build on shared internal tooling, we host everything on our own infrastructure, and it's tied to your Sagan membership.

One process that's repeatable, painful, and specific to how you operate. The best candidates have judgment or content in the loop: screening, drafting, estimating, auditing, flagging. Every build is scoped so the first version is achievable in roughly 10 hours of engineering time. We'd rather ship a focused v1 you actually use, then expand it, than spend months on a monolith.

A pattern we see constantly: the members who get the most out of their agents are the ones who can articulate exactly what they want, with examples. If the rules for your process live in one person's head and have never been written down, the build will stall until they are. We'll help you extract them. One trick we recommend: point Claude at your historical notes or Slack threads and have it draft your rules file. Takes minutes.

A few things:

• Builds that carry a formal compliance regime — HIPAA with a BAA, or anything similar. The compliance overhead runs low six figures on our end, and it doesn't fit this price point.
• Oversight or consulting-only engagements where your team builds and we advise. We own builds end to end. If you want your own team learning to build, that's what Sagan University, the weekly AI Briefings, and the Builder's Group are for.
• Anything illegal, or anything that would breach a contract or a platform's terms of service. Scraping a site that forbids it, working around another vendor's access rules, and so on.

If your project is bigger than a standard build, we won't pretend it fits in one credit. We'll scope it hourly or talk about a dedicated developer arrangement.

Book an AI intake session here: book an intake session

The intake session explores your business and helps define the right first build before anything technical happens. Scoping is free, and you can do as many scoping sessions as you need — the whole point is making sure we're solving the right problem before you spend a credit.

The technical architecture session with our engineering team comes after, and that's where the build begins: it counts against your build credit's 10 hours.

Want to see what other members have built first? Browse examples: spruce-flame-gthx.here.now

The person who actually knows the process. The one who does it today, or will use the agent daily. If the owner books the call but the office manager runs the workflow, bring the office manager. Vague scoping produces vague agents. Also decide up front who gives the green light; approvals stall for weeks when nobody's sure who owns the decision.

Three things, and they're the difference between a two-week build and a two-month one:

1. Real examples. Usually 3–10 samples of the workflow end to end: past projects, scored call recordings, sample invoices, whatever the agent will handle. The agent learns your standards from these, so include the thinking behind them. What you provide is what we build for.
2. Access. See the security section below for exactly what we ask for and how.
3. Your time when we need it. We'll have questions while we build, and we work through them fastest when someone on your side can answer within a day or two. If we're waiting on you for more than five business days, we'll pause the build and pick it back up whenever you're ready.

Builds start at $500, or 1 credit. A credit is worth $500, which buys 10 hours of development time at our standard $50/hour rate. Those 10 hours cover everything from the technical architecture session through delivery, and a standard build is scoped to fit inside them. Bigger builds take more than one credit or get scoped hourly — either way, you'll know the number before you approve anything.

If your membership includes credits, a standard build is covered; if you're out of credits or on a plan without them, you pay cash instead. Whether your plan includes credits depends on your membership tier, so ask your account manager if you're not sure what you have. The credit is deducted when you approve the scope, and every build waits on your written approval before we start.

Once your agent is live, you pay for what it actually uses: LLM tokens, API calls, data, hosting. Ongoing pricing is our cost plus 20%. Most agents run under $100/month. If usage costs more than that, it's typically because the agent is saving a multiple of it elsewhere.

We can't quote your exact number up front because it depends on real usage. You'll see it firm up in your first month or two of invoices.

There's no retainer, no subscription on the build itself, and no seat pricing. If you stop using an agent, your usage bill drops to the floor.

Once your build is delivered, every kind of work on it — bug fixes, tweaks, new features, updating an existing build — is $50/hour with a 1-hour minimum per request. One rate, one rule, whatever the ask.

The minimum is a call-out fee, the same way a tradesperson charges to come out and change a light bulb. Even a five-minute change means an engineer has to stop, reload your project, and get back into it. That switching cost is real, so the hour is what you're buying: direct time with an engineer who knows your build.

Two things that make it go further: we'll estimate the hours before starting so you can approve the number, and we'll batch small requests together so each round earns its minimum. If you have three little tweaks, send them at once.

Bigger additions run as another credit ($500 / 10 more hours) or a separately scoped follow-on build.

This is why the scope email matters so much: it's the record of what your 10 hours buy, and the time to flag a gap is before you approve it.

If we discover mid-build that the work requires substantially more hours than estimated, we flag it with you before doing any additional billable work. We don't run up hours quietly. If you cancel after development has begun, the credit is still deducted, because the engineering time has already been spent.

1. Build. For most builds we start with a working prototype on realistic synthetic data, so you see the full workflow before anything touches your live systems. (Data-infrastructure builds are the exception. Those we build directly against your real systems from the start, with access up front.)
2. Your review. You get the live link, credentials, and a walkthrough.
3. Your response. You have five business days to test and get back to us. If we don't hear from you, the build is treated as complete and paused.
4. Go-live. The agent is yours to run, and usage billing starts.

You'll also have a dedicated project manager as your point of contact through the build and after launch.

We aim to ship within two weeks of scope approval. When it takes longer, the cause is almost always access: expired invites, 2FA, API tiers, a vendor's IT queue. If we hit a blocker, we'll tell you.

Some projects carry real uncertainty: novel data extraction, unproven approaches. We'll say so up front, scope hourly rather than flat, and start with a narrow end-to-end proof. If the approach can't be made reliable, we'll be straight with you and talk through other problems we could solve instead.

It really does depend on the build — that's the honest answer, and it's why we walk through it with you rather than hand you a checklist. The standing patterns:

• An address we give you. We issue a build-specific address for your project and send it to you when the build starts. Don't reuse an address from another build or another vendor: most tools only allow one account per email, so a recycled address collides with an account that already exists somewhere else.
• API keys, read-only wherever possible. Many builds never need write access, and we'll say exactly which scopes we need and why.
• A dedicated mailbox on your domain (e.g., agent@yourcompany.com) for builds that send or receive email. Your team keeps the login and full visibility.
• Sample data exports (CSVs) so the agent starts with full context.

For two-factor systems, we'll suggest a shared authenticator tool rather than texting codes around. All access you grant is revocable by you at any time.

Your agent is a standalone, single-tenant build. Your data lives in its own dedicated environment, never shared or mixed with any other member's. We host on Railway; member agents run at agent.saganpassport.ai. Data passes through LLM APIs (Anthropic, Google) for processing, and neither trains on data sent through their APIs.

What we won't do is negotiate bespoke enterprise legal terms on a $500 build. The economics don't support it. If your procurement process requires that, we're happy to intro you to custom shops that start around $20K.

No. Everything runs in isolated environments inside our infrastructure. That's the reason a build starts at $500 instead of five figures. We connect to your tools via APIs; we don't deploy onto your stack. If self-hosting is a hard requirement, a custom AI shop is the right fit, and we'll make the intro.

When your final build lands, you have five business days to test everything and respond. If we don't hear from you within that window, we'll treat the build as complete and pause it. You can pick it back up with your account manager whenever you're ready. Traveling or slammed? Tell us and we'll work with your timing.

Any fixes, adjustments, or changes after delivery are billed at $50/hour with a 1-hour minimum. So use the window: test hard, test early, and flag everything at once.

Your project manager, on the same email thread as your build, or ai@getsagan.com. Fix work is billed at $50/hour with a 1-hour minimum, and we'll estimate hours before starting so you can approve the number. When you report an issue, separate true bugs from nice-to-haves (it helps us prioritize) and include screenshots or a quick screen recording if you can — clear reports mean fewer billed hours.

Some things, yes: most agents expose settings you control directly. Prompts, screening criteria, confidence thresholds, schedules, phone numbers on outputs. Changes to the agent's logic or flow go through us: small tweaks billed hourly, bigger changes as a new build.

Yes. Apps are configured to your domain, so anyone with your company email can log in. No per-seat charges, ever.

For anything customer-facing (lead replies, SMS), agents ship with safety rails: a confidence threshold (the agent only sends on its own above it; everything else waits in a queue for your approval), a full log of what was sent, and a kill switch. Auto-send stays off until you're comfortable turning it on. SMS builds also require carrier (10DLC) registration, which needs your legal name, EIN, address, and some consent language on your website. We file it for you, but carriers take one to two weeks, so it's often the long pole.

Think of it like a copy machine we put in your office: you pay for the ink, we keep the machine running. Sagan retains the code and operates the runtime (hosting, monitoring, deployments, maintenance). You get unlimited use of the agent for as long as you're a member, your data is always yours, and you'd get an export plus advance notice if a build were ever retired.

We share building blocks between builds, which is why we retain ownership of the IP and why a build starts at $500 instead of $20,000. If you want a custom build owned outright by you, that's a different arrangement — inquire within and we'll walk you through it.

Your agents turn off. Active membership is what covers hosting, maintenance, and support. No membership, no agents. If you're considering a pause, talk to your account manager first so nothing goes dark mid-workflow.

Each agent is hard-coded single-tenant: one company, one use case. Your whole team can use your build, but it can't be copy-pasted to another entity. The path for a franchisee or partner company: they join as a Sagan member and we deploy a clone under the same parameters as yours. There's a referral arrangement in it for you, so ask us for details.

Not directly. Every agent is architected around a specific member's systems, so there's no plug-and-play transfer. What we do instead: scope a version tailored to your business, informed by everything we learned building the original. If you've seen a build you like in the examples gallery or heard about one from another member, that's exactly the thing to bring to your intake session.