Roadmap & versions

The build order for Agentfy.ai 2.0. Each version is a block of work with an explicit scope and an owner (to be filled in). Status legend: 🟢 done · 🟡 in progress · ⚪ planned.

Versions at a glance

Version	Block	Status	Owner
v0.1	Scaffolding — repo, CleanSlice skeleton, infra/setup	⚪ planned	—
v0.2	Core (no-runtime) — agent, llm, chat, memory, orchestrator	⚪ planned	—
v0.3	Ephemeral Light runtime (MVP) — tasks, worker, events, file	⚪ planned	—
v0.4	Browser runtime + billing/usage	⚪ planned	—
v0.5	Heavy + Warm + scheduled	⚪ planned	—
v0.6	Knowledge — LightRAG service	⚪ planned	—
v0.7	Frontend — app + admin + SDK	⚪ planned	—
v1.0	Production — IaC, observability, hardening	⚪ planned	—

v0.1 — Scaffolding

Owner: — · Status: ⚪ planned

NestJS + Prisma + CleanSlice skeleton in api/ (# aliases, per-slice .prisma assembly).
infra (prisma · redis) + setup (core · mcp); wire BullMQ on Redis; health endpoint.
Exit: api boots, migrations run, health is green.

v0.2 — Core (no-runtime)

Owner: — · Status: ⚪ planned

agent (new model), system/llm (Claude gateway, streaming), chat/memory on Postgres.
orchestrator in No-Runtime mode (tool-analysis stubbed → always direct LLM answer).
user (auth — own JWT, team/role/apiKey).
Exit: chat with an agent end-to-end, no pods.

v0.3 — Ephemeral Light runtime (MVP core)

Owner: — · Status: ⚪ planned

runtime — tasks (BullMQ), worker manager (sessions · k8s · idle · profiles), events (SSE/WS).
worker app image — agent loop + bash/fs tools + WS client.
system/file (artifacts/outputs) + usage metering hooks.
Exit: "run a bash script" → Job spins up, streams logs, saves output, autosleeps.

v0.4 — Browser runtime + billing

Owner: — · Status: ⚪ planned

worker browser mode (Playwright + Chromium in-pod; no separate Browserless pool).
billing (subscription/product/price/payment/invoice/webhook) + usage → invoice lines.
Security hardening: NetworkPolicy, short-lived creds, namespace isolation, audit log.
Exit: browser tasks work; usage is metered and billed.

v0.5 — Heavy + Warm + scheduled

Owner: — · Status: ⚪ planned

Heavy profile (dedicated nodepool, high limits) + Warm (premium extended idle).
Scheduled tasks (BullMQ repeatable) → same orchestrator path.

v0.6 — Knowledge (LightRAG)

Owner: — · Status: ⚪ planned

LightRAG service (Python) on Hetzner k8s; agent/knowledge thin gateway.
Per-knowledge-base workspaces; async/queued ingestion (token-budgeted) via tasks.
Storage backend per scale (unified PG pgvector + AGE vs Neo4j + Qdrant). Benchmark first.

v0.7 — Frontend (app + admin + SDK)

Owner: — · Status: ⚪ planned

app (customer cabinet) + admin (panel) on Nuxt; new UIs for runtime sessions + usage/billing.
Embeddable chat widget (SDK).

v1.0 — Production

Owner: — · Status: ⚪ planned

IaC for Hetzner k8s; object storage + datastores provisioned.
Observability (Sentry, logs/metrics), backups/HA, rate-limits.
GA hardening + load test (incl. LightRAG ingestion at target scale).

Team / owners

Area	Owner	Notes
Core / runtime (`api`, `worker`)	—
Knowledge (LightRAG)	—
Frontend (`app`, `admin`, SDK)	—
Infra / DevOps (Hetzner k8s, IaC)	—
Billing / usage	—

Fill in owners as the team forms. Versions map to the phased roadmap in MIGRATION.md and api2/ACTION-PLAN.md.