Inspiration
Every family has a story worth telling, but turning that story into an actual shootable screenplay is still a slow, fragmented process. A biographer, an estate/compliance reviewer, a story editor, a cinematographer, and a line producer all have to be sourced and coordinated separately. Financial and historical sensitivities need governance review before a single page gets written. Pre-production alone can take weeks of back-and-forth before a single frame is ever shot. Our project aims to untangle this system into an autonomous multi-agent system into one continuous, orchestrated run , without losing the iteration and critique that makes a real writers' room's output good in the first place.
What it does
Legacy Chronicles is an autonomous multi-agent production studio that takes a client brief and a family archive as input and produces:
- A cinematic screenplay with full dialogue and scene direction, refined through agent critique loops.
- A visual beat sheet- shot-by-shot visual architecture for the production.
- A production bible (PDF) - a one-click export combining the finished screenplay with a budget-aware production plan.
The whole experience is driven by a single wizard talking to six specialized AI agents, with no manual hand-offs between stages: a Biographer, an Estate Auditor, a Storyline Architect, a Director of Photography, a Line Producer, and a Scriptwriter Studio.
How we built it
The core of the system is a LangGraphstate machine that orchestrates the six agents end-to-end, from the first interview through to the final PDF. This is built as a cyclic state graph rather than a linear pipeline, so agents can critique and revise each other's output, looping between review and rewrite until the screenplay is actually ready.
Each agent has a distinct role:
- Biographer- captures the family narrative from the brief and archive.
- Estate Auditor-runs governance and compliance checks, calling Grafana MCP tools directly at runtime.
- Storyline Architect- builds the underlying story structure.
- Director of Photography—-plans shots and visual style, producing the beat sheet.
- Line Producer- budgets the production using a cost-catalog MCP integration.
- Scriptwriter Studio- finalizes the screenplay and exports the production bible.
The agents' reasoning and generation is powered by Vertex AI Gemini. Every agent call is traced via OpenTelemetry into Grafana Tempo. The Estate Auditor doesn't just log to Grafana, it actively calls Grafana's MCP tools (alert rules, dashboards, Prometheus queries) as part of its compliance decision-making. The backend is served with FastAPI, the client-facing wizard is built in React + Vite, and an MCP server (backend/mcp_server.py) exposes the cost-catalog and Grafana integrations to the agents.
Challenges we ran into
Getting agents to genuinely critique each other's work, rather than just rubber-stamping it, was the hardest part of the orchestration. A cyclic state graph is easy to describe but harder to tune in practice we had to be careful that the review/rewrite loop actually converged toward a finished screenplay instead of looping indefinitely or drifting away from the original brief with each pass. Integrating Grafana MCP as a live, callable tool for the Estate Auditor (rather than just a monitoring target) also meant treating our own observability stack as a dependency the agents relied on mid-generation, which raised the stakes on making those tool calls reliable and well-scoped. Coordinating six distinct agent roles into one coherent output , with no manual hand-offs also meant a lot of iteration on how state and context get passed between agents so nothing gets lost or contradicted along the way.
Accomplishments that we're proud of
We're proud that the Estate Auditor genuinely calls Grafana's MCP tools at runtime rather than just referencing them in a README — governance checks are grounded in live dashboards, alert rules, and Prometheus queries, not just the model's own unchecked judgment. We're also proud of the end-to-end coherence of the system: one wizard, six agents, and a single continuous run that produces a real, usable screenplay, beat sheet, and production bible , collapsing a process that would normally take a human production team weeks into something measured in minutes.
What we learned
The biggest lesson was that an observability stack doesn't have to be a passive, after-the-fact dashboard — wiring Grafana MCP directly into an agent's decision-making let compliance and governance checks happen live, in-line with content generation, rather than as a separate review step afterward. We also learned that cyclic orchestration matters a lot more than we initially expected for creative output quality: letting agents critique and revise each other caught and fixed issues that a single-pass pipeline would have let through untouched. And on the product side, we learned that a single coherent hand-off across agents is often more valuable to the end user than any one agent's individual output quality, the real win is removing the coordination burden entirely, not just automating one piece of it.
What's next for Legacy Chronicles
Next, we'd like to expand the Estate Auditor's governance checks beyond the current compliance scope, add support for longer-form, multi-episode story structures, and let the Director of Photography agent generate actual storyboard imagery alongside the shot list. We're also interested in letting production teams plug in their own cost catalogs and compliance rules via MCP, so the same six-agent studio can adapt to different production companies' real-world constraints rather than a single fixed configuration.
Log in or sign up for Devpost to join the conversation.