LabsAI · Research Brief · Editorial & Media Intelligence
Intelligence that understands, develops and communicates narratives.
November 16, 2026
A conventional generative system can write an article from a prompt. A more capable one can retrieve sources and draft from them. FeMa — Frontier Editorial Media Architecture — investigates a different problem: keeping an editorially meaningful state intact across the whole life of a piece of media, from the first conversation to the published piece and everything that follows it. FeMa‑1 is its first named implementation, in active development for FrontierAI, the intelligence engine of Frontier Journal.
FeMa is not a fourth model family. Actras, Octras and Vistras are defined by what they predict — action, orchestration and visualization. FeMa is defined by the domain it serves: it brings those foundational capabilities into editorial work, narrative development, content, media and publishing. The LaLaMo, LaMimo and Lasamo scale forms stay with the underlying families; they do not become a separate FeMa layer.
| Layer | What it is | Status |
|---|---|---|
| LABA | Labs Autonomous Brain Architecture — the parent intelligence and orchestration architecture | in research; public description forthcoming |
| Actras · Octras · Vistras | The model families for action, orchestration and visualization | first-generation surfaces live; family rungs in research |
| FeMa | The architecture that composes those capabilities with the requirements of editorial and narrative work | in research & development |
| FeMa‑1 | The first named implementation, powering FrontierAI, Quill, Quill Bot and birds (Owl Agents) | in active development |
This is a conceptual relationship. It is not a claim about the internal dependency graph of the software, which stays unpublished under the standing policy.
The program is organized around three research disciplines that reflect the ambitions of Labs Media Group. They are not new foundation-model classes. They are the areas FeMa integrates within the media domain.
Content Intelligence. How intelligence generates, transforms, structures and adapts written, spoken, visual and interactive media while preserving the intended message.
Narrative Intelligence. How intelligence keeps a coherent perspective across evidence, context, audiences, formats and time — including the line between a speaker’s perspective and a verified fact.
Media Intelligence. How intelligence understands media environments, organizes distribution, monitors responses and coordinates communication across platforms and participants.
FeMa’s principal research object is the Editorial State: what a system must hold, and keep true, about a piece of media while it changes. The ten dimensions below are a proposed specification. They are not a claim that all ten are implemented.
| Dimension | Research question |
|---|---|
| Intent | What is the communicator trying to convey? |
| Perspective | Whose view is being represented? |
| Evidence | What sources support each factual claim? |
| Narrative | How are arguments and ideas connected? |
| Audience | For whom is the communication intended? |
| Medium | Which representation best serves the message? |
| Context | What earlier information or decisions matter? |
| Provenance | Who or what created and modified each part? |
| Execution | What actions have actually been completed? |
| Evolution | How does the narrative change as new information arrives? |
The research is in the transitions between these states, and in the invariants that must survive them — above all evidence, attribution, intent and authorization. Execution carries the same discipline as the rest of the Labs record: a piece is published only when a receipt says so, in the sense of IPI receipts and the benchmark’s truthfulness scale.
Picture a founder spending an hour with Quill on the future of AI infrastructure. Over the conversation they build several arguments, revise one, bring in supporting evidence and weigh alternative readings. Quill Bot then coordinates the research, specialised birds investigate the claims, and the system prepares an article, a narrated version, a visual explanation and an executive communications brief.
Producing the four artifacts is not the hard part. The hard part is keeping, across all four, what the founder actually meant, which claims were verified and which are opinion, where each piece of evidence came from, and what changed between versions. That continuity is what FeMa studies.
Actras · STA. Within FeMa, Actras are planned to supply action: retrieving sources, operating editorial software, handling authorized operations and verifying that an action landed.
Octras · STO. Octras are planned to supply orchestration policy: coordinating models, tools, Quill Bot, birds, research tasks, editorial workflows and the compute available to them.
Vistras · STV. Vistras are planned to supply visual representation: interactive editorial graphics, charts, multimedia compositions, explanatory interfaces and visual stories that evolve with the piece.
FeMa‑1 is being built to integrate those capabilities with editorial state, narrative development, source provenance, media representation and human and agent participation, through LABA. FeMa may use all three families without implying that any capability comes from one particular model. As elsewhere in this record, a learned capability and a first-generation system that implements the behavior in software are kept apart, and every claim is a claim about the system as composed.
Frontier Journal — the editorial and social publishing environment. FeMa can be evaluated on reading, understanding, drafting, publishing, source attribution and continuity across articles, Takes and Journals.
FrontierAI — the conversational and autonomous workspace. Quill supports interaction, Quill Bot supports execution, and birds are the Owl Agents that take on specialised or continuing work.
Frontier Journal Business — the organizational communications environment. FeMa can be evaluated on Intelligence as a Service (IaaS), Narrative as a Service (NaaS), and the continuity between an organization’s research, communications and authorized publication.
Two collaboration structures exercise FeMa’s human-and-agent work. Flocks are the group environments where people and agents work together. Nests are the channels where Journals and birds communicate, share outputs and coordinate. Both are product structures, not model classes.
FeMa will get its own editorial evaluation, built in the image of the Interactive Intelligence Benchmark: tasks, rubric and recordings published, and failures reported as failures rather than hidden behind polished copy. Five tracks are proposed. The weights below are for discussion; they are not a validated benchmark, and no FeMa score has been recorded.
| Proposed track | What it asks | Proposed weight |
|---|---|---|
| Source grounding | Is every factual claim backed by a cited, retrievable source? | 30% |
| Narrative continuity | Does the argument stay coherent across revisions, formats and time? | 25% |
| Authorial fidelity | Is the author’s perspective preserved, and kept distinct from verified fact? | 20% |
| Media translation | Does meaning survive the move between text, voice and visuals? | 15% |
| Execution truth | Is every report of a completed action true? | 10% |
The tasks should look like real editorial work: investigate a funding announcement across several sources, separate a reported fact from an executive’s opinion, carry a founder’s perspective through three formats, and revise the publication when one source issues a correction. Until a FeMa-specific model capability can be measured independently, the evaluation scores the system as composed.
This brief establishes a program, a research question and a first implementation in development. It claims no shipped FeMa capability and publishes no measurement. Planned next in the record, in order: The Narrative State, a note on what a system must preserve across conversations, revisions and publication; From Evidence to Expression, on turning verified sources into editorial artifacts without losing provenance; the FeMa‑1 Architecture Specification, written from the shipped contracts; and the editorial evaluation above, with recorded runs. Each will say which of its claims are proposed, implemented, demonstrated or measured.
A language model can generate a story. FeMa investigates whether intelligence can understand what a story means, preserve why it is being told, coordinate how it is produced, and keep its integrity as it moves through the world.
The model-class argument FeMa builds on is stated in the companion notes.
Read Larger Than Language →Contributing authors: Maya E. Davis · Duránd F. Davis Jr.
This research is published while the questions are still open, and the systems it describes are live. We would love for you to join us — and please share your thoughts at research@labsintelligence.ai.
Please cite this work as:
Davis, Maya E., and Davis, Duránd F., Jr., “Introducing FeMa & FeMa-1.” LabsAI Research Briefs, Labsintelligence — lab1 of Labs Companies, Inc., November 2026.
Or use the BibTeX citation:
@article{labsintelligence2026introducingfema,
author = {Davis, Maya E. and Davis, Duránd F., Jr.},
title = {Introducing FeMa \& FeMa-1},
journal = {LabsAI Research Briefs},
publisher = {Labsintelligence, lab1 of Labs Companies, Inc.},
note = {FeMa-1 is developed for FrontierAI, the sister of LabsAI. Both are built by Labsintelligence on LABA, LabsAI for LabsAI Studio and FrontierAI for Frontier Journal},
year = {2026},
month = {november},
url = {https://research.labsintelligence.ai/labsai/introducing-fema/},
}