Roadmap | AgentMargin

Delivered and released

Version	What shipped
v0.1 alpha	Exact cost engine, tenant-scoped ingestion, idempotent usage and revenue records, baseline PostgreSQL tests
v0.2 alpha	One-command local demo, OpenAI usage helpers, unknown-cost diagnostics, live local E2E
v0.3 alpha	Workspace pricing with effective dates and immutable history, React + Arc dark dashboard, local integration testing

Implemented on main, not yet tagged.

Done: Compiled SDK ESM and TypeScript declarations, with verified npm tarball contents.
Done: A runnable Next.js App Router demo that sends synthetic usage and revenue.
Done: Security and onboarding docs; Next.js production build and dependency audit in CI.
Done: Opt-in, operator-reviewed LiteLLM community pricing import with provenance digest.
Done: A weekly, read-only provider pricing candidate scan. It doesn't apply prices automatically.
Done: First public SDK release: @makinohq/agentmargin@0.4.0-alpha.1 on npm under the alpha tag.
Not done: OIDC release verification: publish 0.4.0-alpha.2 through GitHub Actions and verify provenance and a clean install.
Not done: Validate installation with three independent developers.

Done: Server-computed, whole-workspace financial totals with clear unknown-cost semantics.
Done: Tenant-scoped keyset pagination, 1–100 customers per page.
Done: Dashboard KPIs from whole-workspace totals instead of a truncated table.
Done: PostgreSQL tests across 125+ customers, cursor boundaries and cross-workspace totals.
Not done: Audit live provider invoice variances and tiered usage.
Not done: Decide an approval policy before allowing unattended provider imports.

Done: Post-call Anthropic Messages adapter; rejects unsupported cache-creation billing tiers.
Done: Signed Stripe USD invoice.paid conversion to idempotent revenue records (local test receiver).
Not done: Live Stripe sandbox, refund, credit and currency reconciliation, durable retries.
Not done: Expand provider coverage based on real integration requests.

Not done. All of it is mandatory before any production internet deployment.

v0.7. Independent tenant threat model, PostgreSQL least-privilege and RLS assessment, authenticated gateway, TLS and distributed abuse controls.
v0.8. Encrypted backups with restore drills, privacy lifecycle, export and deletion policy, redacted logging, alerts and incident handling.
v0.9. Provider bill reconciliation, refunds and revenue adjustments, migrations with rollback tests, resilience and load scenarios.

v1.0 is tagged only when each gate has independent evidence:

Real developers have installed and exercised the core loop without maintainer help.
The public SDK artifact is verified, released with provenance, and its API contracts are documented.
Tenant security, authentication, data governance and abuse controls are independently reviewed.
Backup restore, migration recovery, telemetry, privacy and incident response are proven by drills.
Provider-billed costs and reconciled revenue match documented accounting rules.

Real developer friction sets the priority. Tell us what broke.