Roadmap posts are easy to write badly. You list a bunch of features in future tense, add some vague dates, and call it a vision. I've read hundreds of them. They're usually marketing dressed up as honesty.
I want to write a different kind of roadmap post — one that actually tells you where things stand, what we're confident about, what's uncertain, and what we're still figuring out. If you're an early adopter or you're evaluating A.L.I.C.E. for your team, you deserve that.
Here's where we are.
The core of A.L.I.C.E. is working. We've built and deployed a 28-agent framework — a team of AI specialists that covers engineering, communication, data, research, security, and more. Each agent has a defined role, a personality, and a domain it owns.
The npx installer makes setup a 2-minute process. You run one command, answer a few prompts, and you have a full multi-agent team ready to go. We support Anthropic, OpenAI, and Ollama — so whether you're paying for Claude or running a local model on your own hardware, A.L.I.C.E. works.
We've shipped Starter and Pro tiers, integrated deeply with OpenClaw as our agent runtime, and built the SOUL/AGENTS/PLAYBOOK memory system — the thing that makes agents actually get better over time rather than resetting to zero on every conversation.
That last one matters more than people realize. Most AI tools are stateless. A.L.I.C.E. isn't. Your agents accumulate knowledge about how you work, what you care about, and what's been tried before. That's not a small thing.
Is everything perfect? No. There are rough edges in the installer on some environments. The memory system works but lacks good tooling for inspecting or editing what agents have learned. And the onboarding experience assumes more technical comfort than it should. We know.
The core product is built for developers. The next phase is about making it accessible to everyone else on your team.
Pro signup flow with browser-based auth — right now, getting on Pro involves more friction than it should. We're replacing that with a clean, web-based auth flow that doesn't require CLI gymnastics.
Persistent cloud memory sync — agents currently store memory locally. That means your memory lives on your laptop. If you switch machines or add a teammate, context doesn't follow. Cloud sync fixes this.
Mission Control web dashboard — a real UI for seeing what your agents are doing, what they've learned, and what tasks are in flight. Think of it as a command center for your AI team.
Team collaboration features — shared agent access, role-based permissions, and the ability to hand tasks between human team members and agents without losing context.
Usage analytics — you should know what your agents are actually doing and what it's costing you. We're building that visibility in.
Q2 is ambitious. Some of this will slip to early Q3. I'm telling you now so it's not a surprise.
By Q3, we want A.L.I.C.E. to be something a 50-person company can adopt without their IT team having a panic attack.
That means Enterprise SSO, shared team memory with proper access controls, and audit logs so you can see exactly what your agents did and when. It means an API for third-party integrations so A.L.I.C.E. can fit into your existing tooling rather than demanding you reorganize around it.
We're also launching an agent marketplace — a place where the community can publish and install specialist agents. The 28-agent framework is what we built. The marketplace is what everyone else builds. The best agents for your industry will come from people who understand that industry better than we do.
The end-of-year goal is simple: anyone should be able to use A.L.I.C.E., not just people who are comfortable with a terminal.
Hosted mode means no local install required. You sign up, you get a team of agents, you start working. Same power, zero setup.
Mobile companion app — most of what people actually need from an AI team (status updates, quick tasks, approvals) can happen from a phone. We're building that experience.
Workflow automation engine — the ability to define repeatable workflows that agents execute automatically, with triggers, conditions, and human checkpoints where you want them.
I'm not promising these dates are fixed. Software is hard, and building multi-agent AI systems is especially hard because the field is moving under our feet. What I am promising is that this is our genuine plan, not a fantasy we reverse-engineered from a pitch deck.
We're building this with early adopters. If you're using A.L.I.C.E. today, your feedback is directly shaping this roadmap. Keep it coming.
If you want to follow along, the best way is to join our early access list or find us where developers hang out. We ship fast and we talk openly about what's working and what isn't.
That's the plan. Let's build it.