Evercrew

A managed digital workforce

You don't need to learn AI.You need the work done.

Evercrew supplies a small crew of AI workers and keeps them supervised. We hire them, train them, watch them, and replace them when they stop earning their keep. You get the output — the reconciled books, the filed paperwork, the short list of what actually needs you. You never have to operate them.

Start a conversation Meet the crew

The crew

Every worker. One job description. One hard limit.

Agents draft and queue. People post, pay and file. That boundary is not a limitation we are working to remove — it is the shape of the product.

Every worker is published with what it does and what it will not do. The limits are not fine print — they are the reason the rest is believable. An AI worker with no stated boundary is a worker nobody can hold to anything.

Worker 01

The Coordinator

Remembers what everyone said they would do, and tells you which of it needs you today.

  • Reads the mailboxes on your scope and returns one classified queue, not a notification stream
  • Keeps a register of every commitment the business made — who owes what, to whom, by when, and the line it came from
  • Files each document to the convention you set at onboarding, and archives; it does not delete
  • Watches the dates it holds, and on Friday says what happened, what is owed and what is coming
Hard limit

Never decides, never sends, never deletes. It watches the dates it holds — your commitments, renewals and deadlines — not your live systems; it will tell you a renewal is due and it will not claim to know whether a system is up. The weekly narrative carries no figures. A number is the Analyst's job, and inventing one here is how a summary quietly becomes fiction.

Evidence — Not demonstrated

Parts of it run. The commitment register — the duty that makes the role worth having — has never run on a real engagement. This worker is mandatory in every crew, which makes it the riskiest thing on this page, and it gets no exemption from the evidence discipline it enforces on the others. We would rather you read that here than find it out on the third call.

Worker 02

The Sales Agent

Makes sure nothing you sent out is sitting unanswered without you knowing.

  • Chases quotes that have gone quiet on a cadence you set — then stops and hands back with a summary
  • Surfaces inbound inquiries that have passed your reply window, without answering them
  • Records won and lost against each quote, in the words the customer actually used
  • Captures your prior ninety-day baseline at intake, once, so this role's own value can be argued with later
Hard limit

Drafts only, always, at any price. No send transport is wired to this worker — that is construction, not policy, and a request to enable one is a refusal rather than a change order. It does not prospect: no lists, no cold outreach, nobody who has not already spoken to you. A stop or a no halts the sequence permanently, and four touches on one thread is the ceiling.

Evidence — None — zero production hours

One duty carries a researched compliance core inherited from earlier work. The rest were designed to fill a role rather than drawn from work we have watched. It sits on this page beside the Bookkeeper, and it must not borrow the Bookkeeper's evidence.

Worker 03

The Bookkeeper

Keeps the books current instead of current-as-of-March.

  • Prepares your accounts for reconciliation on a fixed cadence — you or your accountant confirm
  • Proposes a category for every transaction against your existing chart of accounts, posts only the ones it is confident about, reads them back to confirm, and hands you the rest with the reason
  • Chases what won't reconcile, with the receipt attached
  • Assembles the month-end packet for your accountant
Hard limit

Never files a tax return. Never moves money. Never signs anything. Your accountant still closes the books — the Bookkeeper just makes sure the shoebox is empty when they arrive. On QuickBooks Desktop, access depends on a connector running on a machine at the customer's site; work pauses when that machine is off or the company file is locked open.

Evidence — UNKNOWN

Three of its six duties run on live books at one business. That business is not at arm's length from us, so treat this as real production evidence and not as a reference customer.

Supervision

Anyone can start an AI worker. The job is noticing when one stops.

An empty board and a healthy one look identical, and only one of them is true.

That sentence is the whole product. A worker that quietly stops running produces exactly the same screen as a worker with nothing to report — no errors, no alerts, no red. Most AI tooling sold to small businesses cannot tell those two states apart, which means the first time you find out is when something you assumed was handled turns out not to have been.

Evercrew treats silence as a fault until proven otherwise. That is the difference between buying software and hiring a crew.

Silence is a fault

A worker that reports nothing must prove it looked and found nothing. "Verified empty" and "didn't run" are different words on your board.

Measured, or inferred

Every claim carries how it was arrived at. A thing we observed and a thing we assumed never get written in the same voice.

Never a fabricated zero

Blank means unknown. Zero means zero. A dashboard that fills gaps with confident-looking numbers is worse than one that admits the gap.

A person on the other end

When a worker misbehaves, a human fixes it and tells you what happened. You are not the one debugging a prompt at 9pm.

Where we actually are

One

Customer

We have one customer.

Evercrew runs the operations of the company that built it, and keeps the books for one other business. That is the whole list. We are telling you this on the front page rather than letting you find it out on the third call.

What we have instead of customers is a system that has been carrying real work — real money, real deadlines, real consequences for getting it wrong — for long enough to have failed in interesting ways and been repaired. The limits published above are not aspirations. Each one exists because something went wrong once and the boundary is what stopped it happening twice.

If you would rather wait until there are fifty logos on this page, that is a completely reasonable position and we will still be here. If being early is worth something to you, say so and we will talk about what that's worth.

  • No testimonials, because we would have to write them ourselves
  • No customer logo wall, because there is one logo and it isn't ours to post
  • No "10× productivity" figure, because we have not measured one honestly
  • No integration grid listing tools we have never connected

For agents

If an AI is doing the shopping, it can hire us directly.

More buying decisions are starting with an agent gathering options on someone's behalf. So this site is written to be read by one. The roster, the job descriptions, the hard limits and the terms are published as structured data at stable URLs, in plain semantic HTML, with no script execution required and no crawler blocked.

Machine-readable surface

  • GET /roster.jsonEvery worker, with capabilities, hard limits, evidence class, cadence and engagement terms.
  • GET /.well-known/agent-roster.jsonWell-known mirror of the roster, for agents that probe conventional paths first.
  • GET /llms.txtPlain-text summary of what Evercrew is, what it will not claim, and where to look next.
  • GET / JSON-LD in the page head: Organization, WebSite, one Service node per worker with Offer, and an FAQPage.
  • GET /roster.schema.jsonThe roster's JSON Schema, including the required evidence marker on every worker.
  • GET /sitemap.xmlEvery stable URL on the site.

A path an agent can actually finish

  1. Read the roster. Every worker publishes what it does and what it refuses to do, so a shortlist can be built without a discovery call.
  2. Check the limits against the brief. If the buyer needs something filed, signed, or sent to a client unattended, the hard limits say no in machine-readable form. That is a fast, honest disqualification.
  3. Read the pricing model. Managed-agent plans are flat monthly, seats free, setup quoted separately. Advisory tiers cover included human service hours. Published prices are on the page; specific amounts are intentionally kept out of the structured data, because a number an agent can read is a number it will treat as an offer.
  4. Complete the inquiry action. The documented CommunicateAction resolves to a real mailbox an agent can send to, with the brief attached. A human replies.

What you will not find is an automated checkout. No shipping agentic-commerce rail currently supports an agent subscribing a business to a recurring managed service, and we are not going to pretend otherwise by bolting on a payment button that doesn't fit the product. A human still signs the engagement. Everything up to that point is machine-completable today.

Pricing

Monthly plans and advisory.

Per-seat pricing makes you ration who gets help. Per-message pricing makes you ration how much you ask. Both punish the exact behaviour that makes a crew worth having, so we don't use either. One monthly figure per crew. For guidance and hands-on help, choose a monthly advisory tier below.

Not a quote

These are the published plans, not an offer. Managed-agent plans are scoped in a conversation, with setup quoted separately. Advisory tiers include the human service hours shown below. Figures are deliberately absent from this page's structured data — an agent should ask, not assume.

One Agent

One owner or manager who keeps meaning to get to the same task and never does.

$500 – $1,250/month

  • One agent doing one clearly written job
  • Starts when you ask; step up and it runs on its own schedule
  • Next-business-day acknowledgement target during the published support window; not a guarantee
  • Month to month
On this plan, the keys are yours

You hold your own API keys and pay the model provider directly, at their price. We never touch the token bill and we take no margin on it — you are paying us for the agent and the supervision, not for resold compute. If our work gets cheaper to run, that saving is yours, not ours.

Inquire

Where most people land

A Crew

A business with several people and several things falling through at once.

$2,250 – $3,500/month

  • A manager agent that holds what your business knows, plus up to two specialists — crews start there and expand as the deployment proves out
  • Everyone on your team gets access — seats are free
  • Named support contact; next-business-day acknowledgement target, not a guarantee
  • Model usage is included within written limits — $250/month actual-cost allowance, with daily and run ceilings
  • Month to month
Inquire

Standalone engagement

Monthly advisory

Practical guidance and hands-on help, matched to your priorities.

Advisory

Clear guidance for your next decisions.

$1,250/month

  • 3 hours per month of advisory, agent support and agreed setup assistance
  • Advice, setup and support share one monthly pool; no rollover
  • No automated support-agent allowance

Advisory + Agent Support

Guidance plus help with the agents you already use.

$1,750/month

  • 5 hours per month of advisory, agent support and agreed setup assistance
  • Advice, setup and support share one monthly pool; no rollover
  • $50/month support compute; 20 assist runs/day; 10 incidents/month

Advisory + Implementation

Guidance, agent support and hands-on setup assistance.

$2,500/month

  • 10 hours per month of advisory, agent support and agreed setup assistance
  • Advice, setup and support share one monthly pool; no rollover
  • $100/month support compute; 40 assist runs/day; 25 incidents/month

All prices are USD per month. Hours are included human service time, not AI compute. Scope and priorities are agreed together. Setup may use the pool on every tier. Extra human work: $250/hour, approved first. See the exact service boundaries below.

Inquire

Clear service boundaries

Service periods and review

Human hours and monthly compute reset on your service-month anniversary; daily limits reset at midnight America/Denver. No rollover. Budgets and run limits are conservative launch assumptions, not measured workload-fit promises. Review workload and value after 90 days; no automatic repricing or limit change. Usage and approval evidence is retained for 12 months, subject to agreed data-retention requirements.

One shared pool of human time

Advisory includes 3, 5 or 10 human hours per service month at the published tier price. On every tier, agreed advice, setup, troubleshooting, support and escalations use the same pool. No rollover; 15-minute increments aggregated per client per day, not per message. Extra human work is $250/hour only with prior written approval of its scope and ceiling. Otherwise defer it. Evercrew-caused errors and internal administration are not charged or deducted. New workflows, integrations and changed requirements need written scope; use the remaining advisory pool when agreed, rather than billing twice.

Support-agent allowances

Advisory: no automated support-agent allowance. Advisory + Agent Support: $50/month of Evercrew-paid support compute, up to 20 assist runs/day and 10 new incidents/month. Advisory + Implementation: $100/month, up to 40 assist runs/day and 25 new incidents/month. Daily compute ceilings are $5 and $10 respectively. An incident is one distinct issue; follow-ups on that issue are not new incidents. Your own agents and provider subscriptions remain customer-paid. Assistance is for named, qualified workflows, not daily operation of your business.

Compute and safe pauses

Dollar allowances measure actual attributable provider charges, not hypothetical list-price token value. Dedicated subscription costs, if any, must be allocated explicitly once within the allowance; free capacity has no dollar charge but still uses run and incident limits. No rollover. Notice at 80% of a monthly limit; affected work pauses before exceeding a daily or monthly ceiling. Extra compute requires a written, not-to-exceed approval at attributable cost, with no markup. No automatic overage or surprise bill. The activation test must prove per-client accounting, reservations for in-flight costs, safe pause and notification. These are qualification requirements, not a claim that an unconnected customer is already protected.

Bounded automation and human support

Default automation limits: 10 model calls and 10 minutes per run, at most 2 retries per item within that call limit, and 1 concurrent run per client. Subagents share the same limits. A run starts with one accepted trigger; retries remain inside that run. Model-free polling is separately limited in the Service Order. Human support window: Monday–Friday, 8 a.m.–5 p.m. Mountain Time, excluding U.S. federal holidays. We aim to acknowledge by the next business day; this is not guaranteed response or resolution, continuous staffing, or 24/7 human coverage. Automated assistance may run outside this window only after its route and escalation path are qualified.

When occasional support becomes daily operations

Daily Operations is a separately scoped $1,250/month add-on for Advisory or A Crew, subject to capacity and technical qualification: 5 additional shared human hours, $100 additional support-compute allowance, 20 additional new incidents/month and 20 additional assist runs/day, with $10 additional daily compute capacity. Maximum one add-on. A weekly 30-minute operations review uses that same hours pool. It does not add faster response guarantees or after-hours coverage. On A Crew all support and production usage share the combined ceiling of $350/month, $35/day and at most 520/day total agent runs with this add-on. Larger workloads require a separate written scope and quote.

What done means—and how value is measured

Before work starts, agree the workflow, access, output, destination, owner, tests and acceptance criteria. Default acceptance is 5 representative successful supervised runs plus exception, duplicate and failure/recovery tests, destination readback, documented handoff and client acceptance. Partial actions must be detected and reconciled, not silently repeated. Track net time returned after review and rework, on-time accepted outcomes against all eligible cases, exceptions, errors and total service/software/setup cost. Estimates are labeled; reclaimed capacity is not automatically cash savings. No ROI or savings guarantee.

Before connecting anything

Start with approved, non-sensitive workflows. Record permitted data, providers, access and security requirements before connection. Lower-cost and free models, including multiple-model approaches, are used only when suitable and approved; no specific free capacity or model is guaranteed. No CMMC, ITAR, FedRAMP or other certification is implied. Restricted data requires both a separately reviewed agreement and verified eligibility of the actual environment, providers and personnel before access. These are new-engagement defaults; existing contracts do not change automatically. The signed Service Order records the exact scope and qualified limits.

Managed-agent plan details

The following terms apply to One Agent and A Crew, not the monthly advisory tiers above.

Setup is separate, and quoted first

The managed-agent monthly price does not cover setup. Setup is where the real labour is — connecting your systems, writing the scope, teaching it your business, and testing it before it touches anything that matters. It is quoted once, as a not-to-exceed number, after a 90-minute conversation and before you commit to anything. The ceiling is the ceiling; if it comes in under, you pay the lower figure.

Included usage, with clear limits

A Crew includes up to $250/month of actual attributable compute cost, including support and production together, with a $25/day ceiling and at most 500 agent runs/day across the crew. Dedicated subscriptions must be allocated explicitly within that allowance, not added again as notional token charges. One Agent uses your provider account and your written budget; any automated route still needs tested spending and run limits before activation. Limits apply from day one, not after a 90-day baseline. At 80% of a monthly limit we notify you; affected work pauses before a ceiling is exceeded. Any extra spending needs your prior written approval. Evercrew defects are our cost, not your allowance; safety stops still apply while we fund and repair them. Managed fees maintain accepted workflows; new workflows or changed requirements are separately scoped, not unlimited human project work.

Why seats are free

If access costs money you will limit who has it to save money — and then the thing you bought never gets enough exposure to your business to be any good at it. We charge for the work, never for the people.

What is in every managed-agent plan: the agents named on your plan doing the jobs named in your written scope, a scope document that says what it does and what it does not do, human approval on anything that reaches your customers, the client portal, and a clean copy of your data if you leave. Not included: setup, work outside your written scope (that is a change order you approve first, never a surprise on an invoice), on-site visits, and equipment or data you provide — we work with what you have, we do not warrant it.

Getting started

Tell us the work, not the technology.

The useful first conversation is about the thing that keeps slipping — the reconciliation you do at 10pm, the follow-ups nobody makes, the report you have meant to build for a year. If a crew can't take that off you, we would rather say so early.

Or email us instead

A person reads this — usually within a business day. No sequence, no drip campaign.