What we’re building next
Public roadmap. Updated as we ship. No fake deadlines — quarters are vague on purpose. Every shipped item links to evidence.
In progress
Active work — usually has a branch or PR. Ship windows are estimates.
Windows bridge agent activation
In progressEC2 i-013da1cbb85fd6168 provisioned. Installing .NET 10 SDK, VS Build Tools, HTTP agent. First proven prompt → installable EXE is the milestone.
End-to-end hardening suite
In progress12-prompt regression test running full orchestrate() — currently only parseIntent() runs in hardening. Public proof page at /admin/factory/proof shows green/amber/red per prompt.
SOC 2 Type II readiness
In progressWorking with Vanta/Drata on evidence collection, change control, vendor risk, incident response. Target audit window Q4 2026.
Planned
Committed work — we know how to build it, just haven't started yet.
SSO via SAML + OIDC
PlannedBuilt on Supabase Auth SSO. Per-tenant SSO config at /admin/factory/settings/sso. Force SSO for Enterprise tier.
VPC / self-hosted deployment
Planneddocker-compose stack (Next.js + Supabase + Linux bridge + Windows bridge) + Terraform templates for AWS, GCP, Azure. Runbook at /factory/docs.
Real mobile pipeline (Android APK first)
PlannedCapacitor full pipeline: web shell → Android Studio gradle build → APK. Emulator launch verifier captures screenshot. iOS deferred (needs macOS bridge).
Code editor + file tree in operator
PlannedMonaco editor with per-file syntax highlighting, file tree navigation, diff view vs reference output. "Edit & rebuild" workflow.
Template marketplace
PlannedTemplates auto-populated from successful jobs (status=complete + grade ≥ B). One-click build from template. Per-category browse, versioning, ratings.
Public REST API + webhooks
PlannedPOST /api/v1/factory/jobs with API key auth. Per-tenant webhook subscriptions on job state changes. SDK for Node + Python.
GitHub PR integration
PlannedCommit generated code as a PR to a chosen repo. Useful for teams that want to review AI output before deploy. Optional auto-merge after CI.
Considering
Open questions. We haven't decided. Tell us if you'd pay for it.
macOS bridge node
ConsideringReal Mac mini in a colo + remote-controlled signing for .dmg output. Expensive ($150/mo per mac mini) and notarization complexity. Considering if Enterprise demand justifies.
Multi-agent parallel generation
ConsideringEntity A and Entity B generated simultaneously by different agent instances. Replit Agent shipped this in Feb 2026. We can match it but need to verify it actually improves quality.
Real-time collaboration in operator
ConsideringMultiple users in one interview, cursor presence, edit broadcast. Built on Supabase Realtime. Considering if Team tier customers ask for it.
AI root-cause analysis for failed builds
ConsideringWhen verification fails, the AI auto-writes a one-paragraph RCA in the proof pack: what compiler error, what file, what we tried, why it didn't work. Currently the operator does this manually.
Shipped
Already in production. Click any item to see proof.
Conversational operator with TTS + STT
ShippedInterview chat that produces a manifest via natural-language Q&A. 12 models hot-swappable per message. Voice in, voice out.
Intent engine: prompt → AppManifest
ShippedStructured manifest extraction with 3-retry validation. Capability resolver decides what bridge nodes a build needs.
Durable jobs + artifact registry
Shipped13-state job machine with Supabase persistence. SHA-256 hashed artifacts with 30-day expiry, signed URLs.
Multi-model AI gateway (12 models)
ShippedClaude, GPT-5, Gemini, Mistral — hot-swappable per request. Per-call cost telemetry in operator Trace tab.
Linux bridge with all 14 commands
ShippedSSH-based execution to Runner-1 — exec, file I/O, process control, screenshot, snapshot/rollback. Tested end-to-end.
Public marketing surface
Shipped/factory landing, use-cases (10 industries), compare (5 competitors), pricing, blog, docs hub, showcase, gallery, status, security, enterprise pages. All static + SEO-ready.
Verified compile-fix-launch loop with functional smoke
ShippedRepair agent + launch verifier ran end-to-end on 2026-05-18 (Recipe Manager Web, job 222cdd42). POST + GET functional smoke against the launched server is the pass bar. PNG screenshot capture is parked separately.
Per-tenant audit log export
ShippedCSV / JSONL export shipped at /api/admin/factory/audit/export. Hash-chained for tamper evidence. Records actor_is_agent + agent_run_id columns. Full orchestrator-step emission is still in progress (today: API submissions + admin retries are recorded).
Want to influence the roadmap?
We listen to people who use the product. Email [email protected] or write to us through the operator. Especially if you’d pay for something we’re marking “considering.”