SnapFlow
Settings & Billing
AI provider setup, your credit balance and plan, your profile, account cleanup, and — for founders — the tamper-evident Audit Trail.
Overview
Settings (route /settings) is labeled "Vision Control Layer" in-app — "Configure the planning model, key access, and operator identity that shape how your DevOps swarm reasons and responds." It has up to nine tabs across the top: AI Configuration, Billing, My Profile, Schedules, Skills, Directives, Brain, Data, and — for founders only — Audit Trail.
Roles: who can see what
Every SnapFlow account has one of three roles, shown on the My Profile tab:
| Role | What it means |
|---|---|
| Founder | Full access, including provider connections, the Secrets vault, Audit Trail, and the founder-only Margin Health panel inside Billing. |
| Employee | Normal day-to-day access; cannot connect/disconnect providers, manage service credentials, or see the Audit Trail tab. |
| Client | The most restricted role, for people outside the core team. |
Role-gated controls in SnapFlow don't just grey out for the wrong role — they don't render at all. If you don't see the Audit Trail tab, it's because your account isn't a Founder, not because it's broken.
AI Configuration
"Controls which LLM the Cortex planner uses to decompose DevOps intents into agent DAGs." In plain terms: this is where you tell SnapFlow which AI provider and model powers its planning and its agents.
| Field | What it controls |
|---|---|
| AI Provider | OpenRouter, Claude (Anthropic), OpenAI, or OpenCode (Go) / Custom. All use an OpenAI-compatible API. |
| Base URL | Leave blank to use the provider's default (shown as placeholder text). OpenCode has two possible defaults depending on your plan — the field explains which. |
| Model | A dropdown of models. Click Refresh models (shows "Loading…") to pull your subscribed models live from the provider; until then you'll see a small built-in fallback list. |
| [Provider] API Key | Password field. Already-saved keys show as a masked placeholder of dots — leave blank to keep the existing key. |
| Slack approval webhook | Paste a Slack incoming-webhook URL to get human-in-the-loop approval requests in Slack. Leave blank for the deployment default; clear it to disable Slack approvals entirely. |
| Planning Temperature | A 0–2 slider. Lower keeps the planner deterministic and predictable (recommended 0.2–0.4 for DevOps); the live label reads "Deterministic" (≤0.3), "Balanced", or "Creative" (≥1.2) as you move it. |
Agency Brain Configuration
Four checkboxes (all on by default) plus two number fields, controlling how the swarm uses what it already knows:
| Control | What it does |
|---|---|
| Enable KB retrieval in planning | Whether Company Brain knowledge is pulled into planning. |
| Enable episodic memory retrieval | Whether past run history informs planning. |
| Enable self-healing auto-retry | Whether the swarm attempts automatic recovery from certain failures. |
| Verify task outcomes before trusting them | Whether outcomes are checked against real state before being treated as successful. Turning this off greys out the confidence field below it. |
| Brain Retrieval Top K | How many knowledge entries to pull in (1–20, default 5). |
| Verification Min Confidence | The minimum confidence score (0–1, default 0.65) required to trust a verification result. |
These toggles only control how the brain is used during planning. To actually add knowledge — manually or from a URL — use Company Brain, linked right below the toggles. See Swarm Intelligence → Knowledge.
Click Save AI Settings (reads "Saving…"). A green "Saved" confirmation appears briefly on success; a red message appears on failure.
Billing
Credit balance
"Credits are frozen in escrow before each task and settled only on a hash-verified successful outcome — failed or unverified work is automatically refunded. Monthly plan credits roll over in full; they never expire." Three stat tiles: Lifetime Purchased, Lifetime Spent, and Credit Grants.
Subscription plans
"Premium includes SnapFlow's LLM; BYOK uses your own provider key at a lower monthly rate. Either way, task pricing stays fixed — 1/5/10 credits by complexity, always."
| Plan | Credits & pricing |
|---|---|
| Starter | 100 credits/mo — "Early-stage startups & lean teams." Premium $1,500/mo · BYOK $450/mo. |
| Growth | 400 credits/mo — "Scaling companies shipping code daily." Premium $5,000/mo · BYOK $1,500/mo. The default/featured tier. |
| Scale | 1,000 credits/mo — "Enterprise-grade continuous delivery." Premium $10,000/mo · BYOK $3,200/mo. |
Each plan/mode combination has its own Subscribe button, which shows "Current plan" (disabled) if that's what you're already on. Once subscribed, a "Manage billing" button opens Stripe's billing portal to change or cancel.
Usage by Complexity
"Where your credits go. Simple = 1, Medium = 5, Complex = 10 credits per verified task." Three progress bars — Simple, Medium, Complex — each showing credits spent and task count.
Transaction History
Your 50 most recent credit movements, each labeled by type: Purchase, Grant, Refund, Task charge, Escrow hold, Escrow release — with a timestamp and signed amount.
Founders see an extra panel comparing internal cost-per-credit against a pricing ceiling, for their own reference. It never changes what you're charged — task pricing is always the fixed 1/5/10 credits regardless of what this panel shows. It simply doesn't appear for non-founder accounts.
My Profile
A read-only summary of your identity as SnapFlow has it on record — Full Name, Email Address, and Role (Founder / Employee / Client). "This identity is used for audit context, approvals, and activity attribution across the swarm." To actually change your name, email, or password, SnapFlow points you to the primary agency portal — there are no editable fields on this tab.
Schedules
A pointer card, not an editor: "Scheduled agent runs are managed via pg_cron. Configure cron expressions for automated security audits, cost reviews, and health checks." Click "Manage Schedules →" to open the actual schedule editor, which lives as a panel on the Mission Control home screen.
Skills, Directives, and Brain (shortcuts)
These three tabs are each a single pointer card — the real editors live on the Swarm Intelligence screen:
| Tab | What it links to |
|---|---|
| Skills | "Built-in skills ship with the platform. Custom skills let agencies define their own runbooks as callable JSON-schema tools." → "Manage Skills →" opens Swarm → Skills. |
| Directives | "Behavioral rules injected into the Cortex planner system prompt. Each directive is a numbered instruction the swarm must respect." → "Manage Directives →" opens Swarm → Governance. |
| Brain | "Your agency's private knowledge base — tracks execution history, failure patterns, and manually taught knowledge…" → "Open Company Brain →" opens Swarm → Knowledge. |
Data
"Remove failed and rejected tasks from the database. For completed tasks, only the most recent run per operation is kept."
- Click Clean up.
- Confirm: "This cannot be undone. Continue?" — Yes, delete or Cancel.
- While running: "Cleaning up…"
- On success: a count of how many failed/rejected tasks were removed, plus any old duplicate completed runs.
Audit Trail (Founder only)
A tamper-evident record of every privileged action taken in your agency, visible only to the Founder role.
Chain integrity
SnapFlow re-walks the hash chain covering every recorded event. Chain verified (green) means nothing has been altered; Chain broken (red) names the exact problem. Click Re-verify to re-run the check on demand.
The event list
"Every privileged action recorded by the app, grouped by day. Most recent 200." Filter by outcome (All outcomes / Success / Denied / Error) or by typing into "Filter by action…". Events are grouped under Today/Yesterday/date headers; each row shows the action, an outcome chip, the actor's email (or "system"), and the time — expandable rows reveal the actor's role, IP address, and raw metadata.