The Agent Action Layer

From Context to Action

Less than 0.1% of people truly know how to use AI by aligning Agents with a personal knowledge base. Every day, people generate enormous amounts of Context, but they have no reliable way to capture and organize it.

Memova brings together interactions with the physical world—meetings, files, ideas, images, video, and every exchange with an Agent—inside a private, local, open LLM Wiki. It is not another folder for storing information. It is a personal knowledge layer that a human and all of their authorized Agents can understand and use together.

A Note Produces a Next Step

Granola turns meetings into notes. Memova goes one step further: it turns understanding into action.

A meeting, an AI conversation, or a sudden idea can become a Note in Memova. The Note does not merely record what happened. Memova interprets what the person is trying to accomplish and can directly create what comes next: a calendar event, an email, an HTML Page, research, or content ready to share.

Actions remain connected to evidence and authority. The system distinguishes what is recorded, what is inferred, what it recommends, and what needs approval. A user can inspect the sources behind a proposed action, revise the interpretation, and decide whether the action should occur.

A Note Never Stands Alone

A Note connects with documents, images, videos, historical Context, and the person’s knowledge base. Together, these materials form a Project Book.

Memova does not understand only the materials. It understands the project, the people behind it, the commitments already made, and the result that should be created next. The same Context can become a research Page, a visual story, a project archive, or evidence for the next decision.

Every Page turns Context into structured, shareable knowledge. It joins a living Book, connects to what has already been built, and shapes what Memova creates next.

The Action Loop

The action layer follows a durable loop:

1. Capture a real event as a Note with its source and immediate intent.
2. Connect the Note to the people, project, decisions, and prior materials that give it meaning.
3. Generate an inspectable proposal for the next useful result.
4. Ask for confirmation when expression, permissions, external communication, money, or other consequential actions are involved.
5. Execute the approved step.
6. Write the result and the user’s feedback back into the Project Book.

This writeback matters. Without it, an Agent can act but cannot learn from the outcome. With it, each action becomes evidence that improves future understanding.

Product Architecture Snapshot

Updated 2026-05-20 product model: Inbox & Brief → Knowledge Base → Agent Hands → Outputs, with the whole loop understood as the Memova Second Brain.

1. Inbox & Brief: natural input and alignment layer for notes, meetings, voice, handwriting, thoughts, and agent-ready briefs.
2. Knowledge Base: Obsidian / local Markdown / project memory / structured people, projects, decisions, and commitments.
3. Agent Hands: product-language execution layer; internally the Agent Executor that operates tools, apps, agents, and workflows.
4. Outputs: plans, drafts, calendar events, tasks, code, reports, and follow-ups.

2026-07-19 first sharp wedge: Context → Private Page → Shared HTML / Optional Distribution → Auto-book → Personal Library. This is not a company pivot; it is a concrete, painful founder direction that can become Memova's first entry into user mindshare. Public posting is not required for product value. The private Page remains the source of truth; sharing and distribution are user-confirmed projections. memova_user_manifesto_future_of_being_remembered_2026-07 defines the enduring promise that a person's value is not merely to be recorded but to continue to be understood. The internal v0.5 implementation demonstrates this as one Book with four Chapters and ten human-and-agent-readable Pages. See memova_company_memory_book_v0_5_2026-07-19 and memova_product_book_v0_1_2026-07-19.

2026-08-13 product implication from Garry Tan's “Own Your Intelligence”: preserve the Page-first canonical MVP, but test a lightweight Context Review Skill → Memory Diff → user confirmation → Context Pack → next-task improvement bridge. Skill is an entry surface, Page/Book is a human projection, and the durable asset is the correctly governed canonical context and calibration loop. Personal and company context remain separate authorities; see memova_garry_tan_own_your_intelligence_product_implications_2026-08-13 and memova_context_ownership_portable_skills_2026-08-13.