Roadmap
What we're working on, in priority order.
Direction
AIFactory is an open-source, open-core project. The strategic priority is adoption — getting the self-hostable, review-gated, auditable core into the hands of the engineers who need it most: platform and security teams at organizations that can't send their code to a cloud agent. See Why AIFactory for the positioning, and the GTM strategy memo for the full reasoning.
What that means for the roadmap, concretely:
- The core stays free and open. The pipeline, web UI, multi-provider routing, worktree isolation, and single-tenant self-hosting are the project — and they stay MIT.
- Near-term focus is the adoption path, not more enterprise depth: a frictionless first-run (one-command self-host, local-model support), a verifiable end-to-end demo, and docs that lead with the problem. Enterprise features deepen as real demand pulls them.
- The enterprise edition (multi-tenant, SAML/SCIM, signed audit anchors + evidence export, support) is how the open-source core stays funded — added when organizations ask for it.
Recently shipped
-
Epic #35 — Enterprise v1.1 ✅ 9/9 children shipped (2026-05-28):
- #36 Tenant Isolation Mode (per-org K8s namespace + IRSA + gatekeeper samples)
- #37 gVisor sandbox opt-in
- #38 LiteLLM gateway + audit hook + PII redactor
- #39 Bedrock + Vertex provider support
- #40 S3 workspace storage + Redis pub/sub fan-out
- #41 SAML 2.0 + SCIM 2.0 (security-foundation + routes + helm blocks)
- #42 OpenTelemetry distributed tracing
- #43 Signed audit-chain anchor + ISO 27001 evidence map
- #154 Scoped MCP API keys See the ISO 27001 evidence map + linked concept docs.
-
Epic #92 — Delegation (Copilot + Duo) ✅ — hand the coder phase off to GitHub Copilot or GitLab Duo Workflow while AIFactory keeps the planning + governance. See the Delegation concept page.
-
Epic #82 — Portal-managed Git clones ✅ — clone repos into the portal's workspace root (env-aware on laptop vs Helm PVC), with encrypted-at-rest stored credentials for private repos. See Portal-managed clones.
Shipping now (stacked PRs)
- Epic #35 #41 — SAML 2.0 + SCIM 2.0 (60% landed): security-foundation modules
(replay cache, OneLogin SDK wrapper with XSW + signed-Assertion defences,
HMAC RelayState, SCIM schemas/filters/auth) merged in PR #177; the
external_identitiesschema for multi-IdP linkage merged in PR #178. Remaining: SAML routes, SCIM CRUD routes, identity-provider dropdown, Helmsaml:+scim:blocks. Design doc:docs/plans/2026-05-28-saml-scim-design.md. - #67 — rmux R0b: async wrapper around rmux CLI
- #68 — rmux R1: per-task session lifecycle + WebSocket bridge
- #69 — rmux R2: frontend Live Console tab + Attach UX
- #70 — rmux R3: bundled rmux binary + Helm toggle + dual-image CI
- #71 — rmux R4: Playwright E2E for the Live Console
When the rmux stack lands on dev, the Live Agent Console becomes a first-class feature.
Next quarter
-
Epic #204 — Enterprise v1.2 (planned improvements to v1.1):
- Per-tenant LLM budget enforcement + rate-limiting (Claude SDK wrapper)
litellm.audit.scrubBeforeSendmode (PII removal before LLM vendor)- Per-tenant audit-chain anchor + external publication (S3 WORM / RFC 3161)
- Streaming response audit coverage
- See tracking issue #204 for full scope.
-
Epic #50 — MCP Control-Plane Tools: expose AIFactory itself as an MCP server so Claude Code can create projects, kick off builds, and read QA reports without leaving the terminal.
-
First-class Linear integration: bidirectional sync (today is one-way GitHub import only).
-
Algolia DocSearch on this docs site.
-
Browser-side runtime config so the frontend can talk to a portal at a custom origin without a rebuild.
On the radar
- Per-org rate limiting for shared deployments
- Cost dashboard — aggregate token spend by phase, model, agent profile
- Plan templates — reusable subtask scaffolds for recurring chores (CRUD endpoints, schema migrations, etc.)
- Vendor lockfile review — automated check that your provider mix in
phaseModelsdoesn't lock you to a single vendor
Deferred
- Multi-tenancy beyond per-org scoping (cross-org sharing, etc.) — only if there's pull from real users
- In-browser code editor as primary surface — VS Code already does this well; we focus on agent orchestration
- Hosted SaaS offering — depends on demand; self-hosted Helm is the supported path
How decisions get made
Roadmap changes go through:
- Open a GitHub Issue with the proposal
- Discuss in the issue or in the project's GitHub Discussions
- A maintainer either lands it on the roadmap or closes with explanation
We don't add features that aren't on this roadmap unless they fix a bug or unblock a real user. If you have a use case the roadmap doesn't cover, open an issue first.