All the questions worth asking.

A 20-minute call where we ask you what the task is, what tools it touches, and what success looks like. We send you a written scope within 48 hours. If we can’t take the job on, we tell you fast — and we usually have a recommendation for who can.

Median is 6 days from intake. The longest we’ve shipped in is 15 days — that one needed a custom integration to a legacy ERP. Starter is the fastest (5–7 days); Scale takes longer because there’s more surface area.

A description of the task (one paragraph is fine), 5–10 sample inputs (real emails, real tickets, real whatever), and access credentials to the tools the agent will touch. We’ll specify the exact scopes; you approve them before we wire anything.

No. If you can configure a Zapier zap, you can work with us. We do all the prompt-writing, integration-wiring, and ops. You review the agent in your portal and tell us when it’s right.

Yes. Every agent runs in a sandbox in your portal for at least 48 hours before production. You see the prompts, you replay any test run, you change anything you want. We don’t ship until you say go.

The work is two shapes — an upfront build that takes real time, and an ongoing runtime that we operate. We could roll the setup into a higher monthly but it would obscure what you’re paying for. We’d rather you know.

One end-to-end agent run. For a chat agent: one conversation, however many turns. For a workflow: one execution of the pipeline. We err on the side of one-charge-per-job, not one-charge-per-token.

Yes, any time. We pro-rate the monthly to the day. Switch within 30 days of setup and we waive the difference in setup fees too.

We email you at 80%. Above 100%, we charge the overage rate on your tier ($0.05–$0.15 per interaction). We don’t cut you off; the agent keeps running.

Only if you ask. Default is month-to-month. Annual saves ~17% (one month free) but locks you in — we don’t push it because most operators don’t need it.

Prices are USD, exclusive of VAT/GST. We invoice with your tax ID; you handle the rest.

If we miss our deployment SLA, we refund the setup fee, no questions asked. After deployment, we don’t refund — but you can cancel month-to-month with no penalty.

In an isolated runtime per client, in the US by default (EU residency available on Scale tier). We don’t share data across clients. We don’t train on your traffic. We don’t sell anything.

Yes, every client. The template is on the Security page; we redline if you have your own.

SOC 2 Type I report is in progress; we expect it Q3. Type II will follow ~6 months after. We can share the controls and policies under NDA before then.

Your named operator + an on-call engineer. Access is logged. No one else on the team touches your runtime unless escalated by you.

Every agent ships with input filters and an escalation policy. We monitor for anomalies (sudden volume spikes, unusual tool-call patterns, jailbreak signals). If the agent starts behaving oddly, we kill the deploy and roll back — you get notified, not surprised.

Yes, any time. Logs as JSONL, prompts as TOML, agent config as YAML. There’s an export button in the portal. No “we’ll get back to you”.

Depends on the task. Most production agents run on Claude Haiku, Claude Sonnet, or GPT-4 family. Some specialty agents use open-source models on dedicated infra. We pick the cheapest model that meets the quality bar; we don’t chase the newest.

Yes. Every prompt is version-controlled in your portal — diffs, history, who changed what. You can suggest edits; we apply them or push back.

If it has an API, almost certainly. We’ve shipped agents wired into Notion, Linear, Slack, Stripe, QuickBooks, Salesforce, HubSpot, custom Postgres, Google Workspace, Microsoft 365, GitHub, Zendesk, Intercom — to name a few. If it’s bespoke, we’ll quote on intake.

Yes. Growth includes up to 2; Scale is unlimited. A custom integration is anything beyond our standard library — a private API, a webhook, a CSV drop in S3, an MCP server. We write the adapter, you approve the scopes.

You set the dial. Range is “draft for human review” (the agent prepares; a human sends) → “act and notify” (the agent does the thing; you see the log) → “fully autonomous” (the agent does the thing; you get a weekly summary). Most clients land at “act and notify” for the first month, then graduate.

Every agent has an escalation path — a human inbox, a Slack channel, or a calendar booking. When confidence is low or the task is out of scope, the agent escalates. You can review escalations in the portal and tune over time.

Three layers: (1) automated probes on response latency, error rate, and tool-call success; (2) sampled human review of agent outputs each week; (3) anomaly detection on volume, sentiment, and escalation rate. Issues page you or your operator.

Target by tier: Starter 99.5%, Growth 99.7%, Scale 99.9%. Status page at /status. SLO credits on Scale (1× monthly fee for every full hour of outage below target).

As often as it needs. The review cadence on each tier (quarterly / monthly / weekly) is the minimum. If the agent drifts or the underlying business changes, we ship a fix when it’s needed, no extra charge.

Yes. 30 days’ notice, no penalty. We export everything, archive the runtime, give you the prompts and logs. If you want to start a new agent on the same subscription, we can usually onboard the replacement in under a week.

We hand over the prompts, the config, and the integration code. The runtime itself is ours, but the IP is yours. We charge a one-time handover fee (typically 1× the setup fee) and consult on the transition. Honest conversation — it’s usually more expensive than just staying with us, but we don’t hide the door.

Six people, founded 2024. See the /about page for the team. Operating since spring 2024; ~250 agents shipped to date.

No. We’re profitable on the managed-service revenue. We’ve talked to investors; we’d rather grow at the pace the operations can sustain.

Remote-first. The team is in NYC, Austin, and one in Lisbon. We do an in-person offsite quarterly.

Maybe. We’re picky. See /about#join for the criteria. Short version: you’ve shipped at least one production agent for your own clients, you have references, and you have capacity for at least 3 new clients in your first 90 days.

oxynhq@gmail.com — replies within one business day. For urgent matters as a client, your operator’s Slack or phone (in your portal).