Skip to content

The platform

One customer operating context. Six moments where it gets used.

Follow one change through all of it: PR #1816 moves support-agent from GPT-5 to Claude Sonnet 5. Every part reads and writes the same customer record, so each moment starts from what the last one decided.

For AI support platforms, agent platforms and B2B copilots. It starts in shadow mode: nothing changes in production until you approve it.

Request a reviewSee a customer-impact report

Follow one change through all of it: PR #1816 moves support-agent from GPT-5 to Claude Sonnet 5. Every part reads and writes the same customer record, so each moment starts from what the last one decided.

  1. Before the first changeConnect · ShadowReads traces, evals and the customer record. Changes nothing.63 customers mappedRung: Shadow
  2. Before the mergeCustomer Impactwith Review · TestGives a verdict per customer on the pull request.27 eligible · 14 notRung: Shadow
  3. During the canaryReleaseScopes the canary, and rolls back per customer.Vela rolled back · 26 continueRung: Approved actions
  4. At request timeRuntimeApplies each customer’s policy at request time.Acme: $572 of $600Rung: Approved actions
  5. In the request path · optionalGatewaywith MeterEnforces it exactly, for the workloads that need it.Acme held at $600Rung: Synvolv Gateway
  6. Every monthEconomicswith PortfolioPrices each customer’s margin, and the options.Acme: 41% against 55%Rung: Shadow
  7. Any time, in SlackAsk SynvolvAnswers with its sources, in Slack, GitHub or the app.Why Acme got expensiveRung: Shadow

Under all six

One graph connects customer promises to production outcomes.

Four objects every part reads and writes: each customer’s contract, plan, entitlements and policies as one versioned record, checked before a change ships, not after an escalation.

AI Operating SpecHow your production AI is supposed to operate: workloads, approved models and tools, quality floors, economic thresholds, release rules.
  • support-agent
  • quality floors 0.85 – 0.90
  • margin target 55%
  • releases need an approver
Customer AI PoliciesThe rules for one customer, versioned and approved. Change a rule and the decisions it touches are checked again.

Provider policy · Meridian Health v3 → v4Approved

− Azure OpenAI only, per the BAA+ Azure OpenAI · Anthropic on AWS Bedrock

4 decisions re-evaluated · Release GPT-5 → Claude Sonnet 5: Blocked → Eligible

Production AI GraphWho a change reaches, and why: customer, plan, agent, workflow, configuration, run, cost and outcome.
  1. Acme
  2. Business plan
  3. support-agent v17
  4. refund tool · 3.5 calls
  5. $0.58 per resolution
  6. 41% margin
Decision LedgerEvery material recommendation and action: evidence, approval, action, and expected against actual.
Release
support-agent v17
Approved
Priya Raman
Expected
27 healthy
Actual
26 healthy · Vela rolled back

How far it goes

Start with answers. Add control after the answers earn your trust.

Use it from the app, GitHub, Slack, the CLI, the API and SDKs, OpenTelemetry and MCP.

Three rungs, and the parts on each

  1. 01ShadowCustomer Impact · Economics · Ask Synvolv
  2. 02Approved actionsRelease · Runtime
  3. 03Synvolv GatewayGateway

Find out who it changes before your customers do.

Start with the change your team is already debating. An engineer reviews it against about seven days of your traces.

Request a reviewJoin the private preview

It starts read-only, in shadow mode. Nothing changes in production until a person approves.