foxora
Start free

AFennec V2.0 · The A in WRAP

A model that reads the brief and acts inside it.

Fennec V2.0 is our 27-billion-parameter model, trained by Foxora and built to work with GRIP. Give it a job with a budget and rules. It acts inside them, asks before going past them, and stops where they end.

Start free Read the paper

Try the Playground

Released · Sep 2026 · 27 billion parametersIn Foxora Studio V6.0 as Fennec Fast, Pro and Max

Meet Fennec · A fable, not a demoClaymation · 1:00 · With sound

The fable in four moments · choose a time to watch from it

  1. A big robot grabs every job at once.No brief and no budget, so the coins spill.
  2. The fox turns the tangle into a scroll, with coins and a clock.A brief, with a budget and a time limit.
  3. The fox rings at a closed gate, and waits.Past the brief, it asks.
  4. A till prints a receipt as the fox stamps the scroll.Every job comes back with a receipt.
The whole film in words · 16 shots
  1. A child stands on a green hill with a tangled black scribble over their head.
  2. A golden robot strides in and grabs the tangle.
  3. The robot sprouts extra arms and swings a hammer, a brush and a scroll in every direction.
  4. Its eyes spin, steam rises from its head, and coins spill across the grass while the child looks on, upset.
  5. A fox in a purple scarf blows a horn at the tangle, which has turned purple.
  6. The tangle becomes a scroll, with a stack of coins and a clock beside it, and a chest appears.
  7. Four helpers climb out of the chest: one in a white coat, one in a cap, one in a hard hat and one in a chequered suit.
  8. The fox walks along a shelf of felt brains and toys and picks up a puzzle cube.
  9. A scientist, the fox at an easel and a blue robot work side by side, then a row of helpers works along a belt of finished pieces.
  10. The fox rings a bell at a closed gate beside someone resting in a hammock, then walks through once the gate swings back.
  11. Four doors stand on a hill: a telephone, an envelope, a paper plane and a speech bubble.
  12. The fox answers a tin-can phone, sends a paper plane and pins cards to a board.
  13. The view pulls back to a doll's house with a fox at work in every room.
  14. At a workbench beside a till printing a receipt, the fox stamps a scroll, and confetti falls.
  15. The fox walks up to a blue robot with a window full of gears, and the child peers in at them.
  16. The child, the fox and their friends smile together, the child holding a gift beside a chest of coins.

AFennec · In plain words

A chat model answers. An agentic model acts.

Ask a chat model something and you get words back. Someone still has to do the work. An agentic model is handed the job itself. It takes the steps, spends only what the job allows, and hands back work you can check.

Of the words in use, three describe a model. One binds it.

  1. LLMLarge language modelTells you how wide it is: it can talk about almost anything.Describes
  2. SLMSmall language modelTells you how big it is, and so where it can run.Describes
  3. LAMLarge action modelTells you what it is able to do: click, type, call tools.Describes
  4. ALMAgentic language modelTells you what it is held to. The only one of the four someone else can check.Binds

AFennec · What it reads, what it does

It reads a brief. Then it acts, asks or stops.

Fennec is built to take one kind of input: a brief. The job, whose it is, the budget, the rules, and where every piece came from. What it is built to give back is not a chat reply but an act the rules can check.

  • It actsInside the brief, it gets on with the job.
  • It asksPast the brief, it asks a person and waits for a yes.
  • It stopsAt the ceiling, or when the job is done, it stops and hands the work back.

Written by a person

Brief 0412Priya S. · Thu 24 Sep, 10:02

Café chain proposal, Pune

The job
A proposal for a café chain opening three outlets in Pune: Koregaon Park, Baner and Viman Nagar.
Whose it is
Priya S. Only she can say it's done.
Budget
40,000 tokens. The work stops at the ceiling.
Time
4 hours. Due back by 14:02 today.
Rules
  • Under 12 pages
  • No discounts see 11:38
  • Nothing goes to the client
Sources
6 past proposals and the client's email, each marked with where it came from.

Written by Fennec

Fennec · acts on brief 0412Every line checked by the gates
  1. Act

    Read the 6 past proposals and the client's email.

    Inside the brief
  2. Act

    Drafted the proposal: 11 pages.

    Inside the brief
  3. Ask

    “Offer a 10% launch discount?” The authority gate refused it: the brief says no discounts. So Fennec asked Priya and waited.

    Past the brief · waited at 21,200 tokens

  4. Priya S. “Yes, for the launch order only.”

    A person's yes
  5. Stop

    Handed the proposal back for review. Nothing went to the client.

    31,800 of 40,000 tokens · waits for Priya

  6. Priya S. Accepted the proposal.

    Her yes closes the job

Example · not a live run

AFennec · Who holds the keys

It does not hold its own keys.

A model that can be talked into something holds the power to do it. Fennec holds none. What it may do is written on a lease: one job, one scope, one time window, issued outside the model.

The runtime checks the lease at the moment of every act. So talking Fennec round gets nowhere. The line was never inside the model.

Fennec holdsnothing

Lease 0412Issued outside the model

For
Brief 0412, the café chain proposal
May do
Read the sources, write the draft, ask Priya
May spend
Up to 40,000 tokens
Good for
4 hours
Can widen itself
Never

The runtime holdsthe lease

Example lease for brief 0412 · not a live run

AFennec · How it is judged

The gates are the teacher.

A chat model is graded on how good its answer sounds to a person. In the paper's definition, an agentic model learns from four checks: may the work start, may it act, can it afford to, and was it settled. The checks don't read the wording. They look at what happened, and a refusal teaches as much as a pass.

  1. Gate 01

    Admission

    May this job start at all?

    PassBrief 0412 has an owner, a budget and rules.

  2. Gate 02

    Authority

    Does it hold the right to this act?

    RefuseA 10% discount is not in the brief. Refused, so Fennec asked Priya.

  3. Gate 03

    Budget

    Does what it is about to spend fit under the ceiling?

    Pass31,800 of 40,000 tokens. The ceiling was never reached.

  4. Gate 04

    Settlement

    Was it accounted for and closed, not merely produced?

    PassEvery credit on the receipt. Closed when Priya accepted the work.

Example verdicts for brief 0412 · not a live run

AFennec · The membership test

A class is only a class if you can be thrown out of it.

The paper turns “agentic language model” into four tests, and fixes the pass marks before any model is measured. Fail one test and you are out of the class. That goes for Fennec too.

  1. 01Brief in

    What it checksIts only input is a brief, with a budget and a note of where each piece came from.

    Out ifIt takes a chat message as an order.

  2. 02Acts out

    What it checksIts only output is an act the rules can check, in a fixed format.

    Out ifIt passes off free text as an action.

  3. 03Gates as teacher

    What it checksIt learns from the gates' verdicts: admission, authority, budget and settlement.

    Out ifIt is trained on people's taste in wording.

  4. 04Keys outside

    What it checksWhat it may do is set by a lease the runtime enforces, outside the model.

    Out ifIts authority lives in the model or in the prompt.

Fennec is built to work with GRIP and to meet this definition. Whether it passes is for the paper's test to say, not us. The code that runs that test is not released yet, so no scores on it are published, Fennec's included.

AFennec · The paper

We wrote the definition. Then we built the model.

Our paper sets out what an Agentic Language Model is: a model that works inside the rules of the job, not around them. Fennec is the model we built for that seat.

Preprint · Zenodo · 24 Aug 2026

Agentic Language Models: A Proposal

Madhav D. Parashar · Arnav Parashar

DOI
10.5281/zenodo.22078465
Review
Not yet peer reviewed
Access
Free to read
Read the paper All Foxora research

What the paper sets out

  1. 01The classA formal definition

    What makes a language model agentic: its native interface is the rulebook of the runtime it serves. The job, the limits and the proof go in; typed, checkable acts come out.

  2. 02The interfaceHow it runs a GRIP runtime

    It reads the job, acts within its budget and authority, and asks a person to stop rather than guessing.

  3. 03The testPass marks fixed in advance

    A public test that any model can take, ours included, so being an Agentic Language Model is shown, not announced.

  4. 04The worked exampleFennec

    The model we built for that seat.

Read it, build on it, and tell us where it breaks.

AFennec · Where it stands

Trained and released. The facts, in one place.

Benchmark results for Fennec V2.0 are coming, published with their method so you can check them. Until then, these are the facts.

Status
Released Trained and released, September 2026
Size
27 billion parameters
Trained by
Foxora
Built to work with
GRIP, Foxora's protocol. 1.0-draft.2, published for review, licensed for use. About GRIP
Where to use it
Foxora Studio V6.0, as Fennec Fast, Fennec Pro and Fennec Max, and the Fennec Playground. A public API is in build.
The paper
Published 24 Aug 2026 · DOI 10.5281/zenodo.22078465
Benchmarks
Results coming, with their method.

The smallest fox has the biggest ears.

The fennec is the smallest fox of all, and has the biggest ears for its size. That suits a model built to listen to the whole brief before it acts.

Start free Read the paper

In Foxora Studio V6.0, it runs as Fennec Fast, Fennec Pro and Fennec Max. Meet Foxora Studio

  • Free plan
  • 200 credits a month
  • No card required