Fabric | Work-Centred AI Platform

Choose the audience that matches your question

The page contains three selectable readings of the same subject.

Analyzed request, identified 3 key concepts

Retrieved context from the Dweve Loom knowledge base

F(n) rounds phi to the n over root five.

Ratios of terms approach phi, near 1.618.

The sequence appears across nature and tiling.

Iterative growth, linear in n, exact in integers.

Dweve Loom, Query Analysis, 3 domain specialists, 2 sources

Type a message or @ to mention an agent...

Coding agent and operator assistance. Demo metrics are illustrative.

Terminal, search, lint, test, git, and more.

Remembers your codebase and team context.

Specialised agents collaborate across concerns.

Review diffs, request changes, final sign-off.

Replay any session bit-for-bit when something needs review.

Autonomous agents that write code and keep receipts

to handle network timeouts, 5xx responses, and idempotent-safe conditions. Pure helper, fully unit-tested.

Seven concerns define an agent, and one of them changed

a person, carrying the roles their account holds

a workspace-scoped token, carrying exactly the same roles

one work item, one event type, one order

the same work item, advanced from two clients

the transition arrives at the subscriber

the work view renders a state it did not write

two writes, two reads, two events, one order

A token is held to what the workspace already sets for a person, so nothing routes around governance by leaving the browser.

Business Workspace from Starter upward, and every licensed deployment. Not in the consumer plans.

The interface reads the timeline it writes to

created, advanced, watched and rendered back

No client holds a state the others cannot see

the model slot moves from external to local to Loom

a team card keeps its roles while one member's model changes

roles kept while one member's model changes

Work graph, roles, records, and concepts.

availability differs, the graph does not

model and integration availability changes per boundary

Each keeps its own boundary, guarantees, and evidence depth.

Context, sources, and decisions never fragment between them.

Direct computation in strategic deployments.

a horizontal work spine gathers the vertical stack

external models and integrations enter from the side under explicit boundaries

Replace any product in the stack. The spine keeps its context, its sources and its signed record exactly where they are.

each row shows status, owner, version, and active work references

every extension connects to a work graph and activity stream

Typed ports catch shape mismatches before the flow runs.

Branches, retries, and human gates stay explicit in the graph.

one participant receives a scoped context envelope, not the whole room

a synthesis card is visibly built from selected messages

receives only clause 7 and policy KB-114

one model swaps out for another without the work card moving

several participants draw different scoped projections from the same graph

earlier work stays connected to superseded versions

separate contributions merge into one result through a named rule

Inference, training, fine-tune, region, tools, and trace depth each carry a value.

The differences stay in policy and evidence, never flattened away.

each lane badges inference, training, fine-tuning, region, tools, and trace depth

Chat, memory, tools, agents, logs, and region move with it.

Knowledge, agents, flows, roles, and evidence stay in place.

chat, memory, tools, workflows, logs, budget, and residency pulled into orbit

the work object stays central, the model sits on one replaceable orbit

replacing it becomes a platform migration

a failed evaluation takes a visible non-production branch

a model budget nears its limit and an audit query resolves historical access

a closed external model appears as a clearly labelled opaque boundary

The same thread follows the work into each form.

Pending decisions travel to the next form.

People and models keep their roles across forms.

the same work ID and source bundle persist across every form

General Reasoning Expert, 85%, 256 tokens, 968ms

clarity Clear, 5 words, complexity Surface

The boundary runs through the card, not around it

the provider explanation sits beside this card as provider output, attributed and unpromoted

Dweve Loom · pattern-matching · 1 expert · 30 sources

sent and returned, timed by Fabric; token count as the provider states it

Four fields unavailable, named rather than dropped

Session-Context, 75% relevance, 1 document

A provider explaining its own work is filed as provider output, beside the record and attributed. It never crosses into the evidence Fabric owns.

Read it as a statement, kept with the source that made it.

Read it as a measurement Fabric took itself.

Read the gap as a gap. Fabric does not close it for you.

choose Fabric Personal or Friends & Family

each connection feeds one task, not separate apps

personal data managed through the public Dweve Mesh

a small hosting map plus a local-device boundary

personal data is hosted on European soil

kept on your own device where the help allows it

organisations can choose private or offline hosting

the first answer can become any of the four

keep it, hand it on, save it, or share it

everything you do stays part of the same task

one synthesis card says exactly which messages it used

a team drawer opens beneath for harder work

give it a purpose, its knowledge, and clear limits

it is ready the next time a similar task comes up

hand a trusted helper to people you choose

all of it attaches to the same task card

can you plan next week and keep it all together?

done, everything below stays on one task

add files and invite the right kind of AI

the answer, its source, and your steps stay on one task

each shows a simple label for what it is good at, not technical detail

you see what each is good at, not technical detail

it changes, the old answer keeps its version

sorted into your library, not lost inside one chat

find it again by name or by what it says inside

before, you explain the background again

after, each helper gets the part it needs

different helpers enter and leave without duplicating the folder

the topic it matched and how clear your question was

you can read the answer normally, the record is there when it matters

renew before the 14th, the earlier rate still applies

the tabs show the files, help, and choice behind it

only what really happened, put in plain terms

the paid route is EUR 19 a month, cancel any time

after the first paid month, eligible customers may use the published spare-compute route under its current terms

collects the files and details the task needs

you check the result before anything is sent

next month, with a new file, the very same way

an excluded private file stack stays behind

the answer, its source, and the useful thread

they can open it, not just trust a picture

your other private files never leave with it

Recorded principles: transparency, reliability, explainability

Scores are recorded per run, so this page does not quote one.

Every candidate stays on the record, not only the one that merged

Counts are per run. On the short run only the first band is filled.

a team object links several agents through Nexus-backed roles

each node writes a different artefact type

a closed-model judge is recorded as non-proof

parallel work merges through a declared rule, not prompt convention

each node links to the underlying object and event

each envelope lists included and excluded nodes with reasons

unsupported external nodes are removed or outlined per posture

each adapter carries a different metadata envelope

manifest fields, version, permission scope, and emitted events

one model node hot-swaps without altering upstream or downstream contracts

one provider disappears and a fallback replaces it without changing the work ID

the registry receives only models that pass the publication gate

eligibility filters visibly remove models for one regulated task

answered from versioned role state and events, not today's configuration

each participant carries a different permission and context scope

Fabric provides the work-centred object model and surface

one task contract is filtered through policy gates before one model is selected

all four write and read the same event types

the boundary carries known metadata and an explicit list of unavailable internals

each stage appends structured state without changing identity

Every entity resolves to an owner, a version, a state, and live work references.

hover detail shows owner, version, state, and work references

The first attempt timed out. The error boundary held and the second attempt returned.

A named person has to sign before the last node runs.

The action runs on the signature, not on the answer.

Every node emits into the same work record

opens to the same depth as a single answer

Governance is not a page beside the work. It is attached to the work, at five points.

used inside that scope, and nowhere else

A question about a past quarter reads the policy and role versions that were in force then, rather than inferring an answer from the configuration standing today.

Every one of these is a stored artefact, not a reconstruction

five artefacts attached to one piece of work