Rally

Inspiration

You're in a group chat with your friends trying to plan a camping trip. No one can figure out when everyone is free. You get lazy on planning because people aren't responding. You realized you forgot to call your parents yesterday. You haven't talked to your best friend from high school in six months.

Relationships can sometimes get messy. But it's not your fault. That's why we built Rally.

What it does

Rally is an iMessage assistant that helps friends and families turn good intentions into time together. Call it into a group chat, and it catches up on the thread, remembers that group’s context, and suggests a useful next step when plans stall. It also supports a shared group history and personal reminders to help important relationships stay on your radar. Rally knows when to stay quiet and asks for approval before taking action, leaving the people in the conversation in control.

How we built it

We built Rally around iMessage, where many plans and relationships already live. A Python and FastAPI service receives messages through BlueBubbles, catches up on the group thread, and stores plans, group memory, reminders, and delivery state in a database. Grok and Grok Voice help interpret conversations and draft replies, while Rally’s code handles group boundaries, duplicate messages, safety checks, expiring proposals, and approvals. A companion web experience provides group history and relationship context, but the conversation remains the main interface.

How AI is Essential

We use a lightweight agent system with access to browser, calendar, events, etc. in order to manage relationships beyond just textual understanding. Grok interprets a messy, ongoing conversation: who is available, what people ruled out, and what still needs an answer. Rally keeps each group’s context separate and checks approvals in its own code before taking an action. Grok Voice makes that same context available through a conversation you can have out loud.

Challenges we ran into

Keeping replies quick took more than choosing a faster model. Model calls, iMessage delivery, and work waiting behind other messages could all add delay. We profiled the path from webhook to reply, reduced the context sent for routine answers, and moved heavier extraction and memory learning out of the immediate reply path.

We also had to make Rally helpful without letting a model invent facts, speak for the group, or act beyond what someone approved. This challenge required a balance between safety and quality. We implementing guardrails and fine-tuned the agent to ensure it kept responses relevant.

Accomplishments that we’re proud of

Rally lives in the conversation. Friends can call Rally in their existing iMessage group instead of moving everyone to another app. On its first call, Rally catches up on recent messages, so it can help with a plan that started before anyone thought to ask it.
It turns messy group talk into a next step. “Dinner Friday,” “after seven,” and “anything but sushi” become a grounded picture of what the group wants and what is still undecided. AI makes sense of those scattered messages; Rally can then suggest one concrete way forward. Our prototype carries that flow through explicit approval to a clearly labeled reservation and a reply in the same thread.
It helps relationships last beyond one event. Rally keeps group-specific context, supports a shared history, and offers personal reminders for people you want to stay close to. The point is more dinners, check-ins, and time together—not another task for the group organizer.
It knows when to step back. Rally can follow a conversation without answering every message, lets people forget stored group facts, and requires approval before taking action. That balance lets AI handle coordination while friends keep control of their plans and relationships.

What we learned

A useful conversational assistant needs good timing and judgment as well as good answers. We learned to measure the whole journey from incoming message to delivered reply, and to keep slower background work from delaying the conversation. We also learned that trust comes from grounding suggestions in what people actually said, making memory and boundaries clear, and keeping people in control of actions.

What’s next for Rally

We want to connect Rally to more of the services people need to turn a plan into an event: scheduling, appointments, reservations, party planning, and catering. We also want to deepen the relationship dashboard and connect reminders, availability, and reviewed actions into a smoother experience. As those integrations grow, Rally will be more interconnected with how you interact with others and save you additional time.