Reference
Glossary
Every SnapFlow-specific term used in this manual, defined in plain English, alphabetically.
Terms used throughout this manual
Plain-English definitions of every SnapFlow-specific word or phrase used elsewhere in this documentation, in alphabetical order.
| Term | Meaning |
|---|---|
| Agent | One specialist AI worker in the swarm — e.g. Terraform (provisioning), Sentinel (security), Ledger (cost). See Swarm Intelligence → Fleet. |
| Blast radius | The list of other resources that would be affected by a proposed change, shown on an approval's Gate Evidence card. |
| Boundary / Environment | A named deploy target (e.g. "staging", "production") with its own approval policy and autonomy limits. See Infrastructure & Targets → Boundaries. |
| Change-as-PR | An optional setting that makes SnapFlow open a reviewable pull request (with Terraform import blocks) for every provisioned change, instead of only changing things directly. |
| Company Brain | Your agency's private, accumulating knowledge base — things you've taught it and lessons it has learned from past work. See Swarm Intelligence → Knowledge. |
| Credits | SnapFlow's billing unit. Tasks cost a fixed 1 (simple), 5 (medium), or 10 (complex) credits regardless of how long they take. See Settings & Billing → Billing. |
| Critical floor | The fixed set of actions — deletes, renames/moves, permission changes, anything in production — that always require your approval, no matter which permission mode is active. See Approvals & permission modes. |
| Directive | A standing instruction you give the planner that applies to every mission, e.g. "Always create a rollback plan before production changes." See Swarm Intelligence → Governance. |
| Escrow | Credits are held ("frozen in escrow") when a mission starts and only actually charged once the outcome is verified; failed or unverified work is refunded automatically. |
| Flight log / Live activity | The real-time, scrolling feed of what the swarm is doing during a mission, with a Pretty (readable) and Verbose (raw) view. See Mission Control → Live activity. |
| Founder / Employee / Client | The three account roles. Founder has full access including provider connections and the Audit Trail; Employee is normal day-to-day access; Client is the most restricted. See Settings & Billing → Roles. |
| Gate / Gated | A step that has been flagged to pause and wait for a human decision before it runs. |
| Engagement | A mission you hand to SnapFlow as a Forward-Deployed Engineer service rather than driving yourself — a migration, embedded operations on one of your servers, or a custom build. Priced at a premium, but settled under the same verified-outcome rule. See Forward-Deployed Engineer. |
| Mission | One end-to-end request you give SnapFlow, from your plain-language intent through planning, execution, and a verified outcome. See Mission Control. |
| Monitor | A recurring, plain-language health check that watches part of your infrastructure and raises an incident in Triage when it finds a problem. See Triage & Gates → Monitors. |
| Permission mode | The agency-wide autonomy setting — Plan only, Guarded, Autonomous, or Unattended — that decides what the swarm can do without asking first. See Approvals & permission modes. |
| Policy | A guardrail rule your team writes that blocks or gates matching actions automatically. See Swarm Intelligence → Governance. |
| Postmortem | An auto-drafted write-up of a resolved incident — summary, impact, hypothesized root cause, remediation, and action items. See Triage & Gates → Postmortems. |
| Pre-Flight Manifest | The confirmation screen shown before any request is dispatched, summarizing risk, cost, and what will happen. See Mission Control → The Pre-Flight Manifest. |
| Sandbox | The public, no-login demo at /sandbox where anyone can try SnapFlow on a pasted log or a public repo. See Getting Started → Sandbox. |
| Shareable report | A public, read-only link to one task's result, meant for pasting into Slack/Teams. See Mission Control → Sharing a result. |
| Simulated mode | What SnapFlow runs in when no real cloud provider is connected — results are illustrative estimates, not real infrastructure changes, and cost fewer credits. |
| Skill | A reusable runbook (or external tool call) that gets injected into a matching step when its trigger words and agent match. See Swarm Intelligence → Skills. |
| Swarm | SnapFlow's collective term for its fleet of specialist AI agents working together on your infrastructure. |
| Task | One step of a mission, carried out by a single agent. See Mission Control → Task detail. |
| Triage | The screen listing everything currently waiting on a human decision — approvals and open incidents together. See Triage & Gates. |
| Verified / Unverified | Whether SnapFlow's automated grader confirmed a mission's outcome against real state. Unverified results pause for you to confirm before being charged. |
SnapFlow