case-study

Momenta OS

An AI OS harness that syncs across the whole team's ChatGPT instances — one shared vault of client context, brand voice, and institutional memory.

  • ChatGPT Agents
  • Knowledge Vault
  • Prompt Systems
  • Workflow Design
  • Feedback Loops
Momenta OS cover

Problem & user context

Marketing teams managing LinkedIn for multiple clients burn hours per post: researching each client’s audience, matching their brand voice, drafting copy, and routing everything through approvals. Worse, every teammate’s AI assistant starts from zero — context lives in scattered docs, and two people asking ChatGPT for “a post for the same client” get two different brands.

Momenta OS fixes the sync problem. It’s an AI OS harness distributed to the whole Momenta Fire team through a shared ChatGPT project/agent config, all pointing at one knowledge vault: client folders, brand voice docs, content rules, approved and rejected examples, feedback logs, and automation docs. The vault is the single source of truth; every teammate’s assistant behaves like the same assistant.

How it works

Every request follows the same workflow, enforced by the harness:

  1. Load the right context. The assistant pulls from the vault structure — the client’s folder, brand voice doc, content rules, approved/rejected examples, and feedback history — before writing a word.
  2. Follow the right workflow. For LinkedIn content: identify the client, load context, review examples and feedback, draft the post, score it against the brand rules, revise, then hand back a decision point — revise, approve, reject, or mark as published.
  3. Respect the feedback system. One offhand comment doesn’t silently rewrite brand rules. Repeatable feedback gets turned into a proposed update that goes through an explicit update process — institutional memory changes deliberately, not accidentally.
  4. Create usable outputs. Posts, visual briefs, captions, client onboarding docs, content calendars, feedback logs, and vault organization — each filed where it belongs.

Day-to-day usage is command-like: “create post for ChangeOvr about filter maintenance”, “score this post”, “log this as approved feedback”, “help onboard a new client”.

Constraints & tradeoffs

  • Harness over app. Instead of building custom software, the OS lives inside the tool the team already uses (ChatGPT) via project/agent config — zero onboarding friction, instant sync when the config or vault updates.
  • Vault as source of truth. All client knowledge is externalized into the vault rather than living in any one person’s chat history — the team can grow without losing memory.
  • Governed change. The scoring and feedback-approval loop trades a little speed for brand consistency clients can trust.

Architecture

Team member A ─┐
Team member B ─┼─► ChatGPT (shared project / agent config = Momenta OS harness)
Team member C ─┘            │
                            ▼
                  Momenta OS knowledge vault
                  ├── client folders (context, onboarding)
                  ├── brand voice docs + content rules
                  ├── approved / rejected examples
                  ├── feedback logs ──► proposed rule updates (governed)
                  └── automation & workflow docs
                            │
                            ▼
        Drafts → score → revise → approve / reject / published

Results & lessons

  • The whole team’s AI now speaks each client’s brand voice consistently — no per-person prompt drift.
  • Feedback compounds: every approval/rejection makes the next draft better for everyone, not just the person who received the note.
  • Lesson: the vault structure matters more than the prompts — a well-organized source of truth makes even simple instructions reliable.