Kevin Hu 82c76a82a9
feat(knowledge): read every format, show what the index holds, and let an agent search it (#489)
<img width="2318" height="1448" alt="image"
src="https://github.com/user-attachments/assets/bca71d69-4972-420c-b14e-c8cf74a17de9"
/>

## Summary

Everything a knowledge base could not read, and the pieces it cut them
into
saying what they are.

**Reading a PDF.** A PDF records glyphs at coordinates; whether they are
a
heading, a table cell or a running footer is inferred from how they
look. There
are now two readers. The geometric one infers structure from type size
and
position and needs nothing but the file. The other asks deepdoc's
models, ported
into `raven/knowledge/parser/deepdoc`: a layout model that says what
each region
is, a detector and recognizer that read a page carrying no text at all,
and a
table-structure model that rebuilds a grid from a picture of one. 
 Detecting boxes around text the file already spells
out exactly costs an inference pass per page to learn less, and
recognising it
costs accuracy.

The layout graph answers with a shape its own postprocessing
raises on, and its classes are DocLayout-YOLO's order rather than the
list
Both were established by running the model, and the label order is
pinned by a test.

**Tables go in as tables.** `<table>` is the only shape that keeps what
a grid
states -- which column a cell stood under, which cells are merged, which
row is
the header -- and it is the shape a model reads a table in, which is
what these
chunks are for. PDF, Word and PowerPoint all state their merges exactly,
and
they now survive as `colspan` and `rowspan`. The flat form this replaces
had to
repeat a header standing over two columns across both of them, because
there was
nowhere to record a span; it had no callers left afterwards and is gone.
Two
deviations from upstream, both about what the string is for: every
cell's text
is escaped, where upstream interpolates it raw and a cell holding a `<`
produces
markup that is no longer a table, and the span attributes are written
only where
there is a span, where upstream emits `<td  >` for every ordinary cell.

**A picture of where each chunk came from.** A chunk is a few hundred
words with
a page number on it, which is enough to cite and not enough to answer
whether
the parser read the page correctly. So every PDF chunk that knows where
it sits
carries one image: the regions it covers, cropped out of the rendered
pages and
stacked. here they are already structured in the parser's element spans,
so
the markers would only put text nobody wrote into the thing being
embedded.
The panel draws a chunk's tables as tables, parsed and
rebuilt as elements rather than injected: a person can rewrite any piece
from
that same panel, so what comes back from the store is user input however
it got
there.

**The page that reads them.** Upstream retired the knowledge page while
this
branch was out; it is rebuilt against the new architecture, and the half
that
never existed is the one a reader needs -- what a document actually
became once
it was cut and embedded. Documents sit in one-level folders and arrive
by four
routes: a file, a folder, a URL, a written note. Adding and indexing are
two
calls on the contract and nothing drains that queue on its own, so a
page that
stopped after the first left every upload saying Queued for good.

Opening a file takes the whole page: the document on the left, the
pieces the
index holds of it on the right. The left half is the file drawn as its
own
pages rather than framed by the browser's PDF viewer, which reads where
to open
once and answers to nothing afterwards -- and which nothing can be drawn
on.
Pages are rendered once and kept, and the region a piece was cut from is
washed
over the page it came from, placed in fractions of the page so the
picture's
own scale never enters the arithmetic. Clicking a piece takes the file
there;
double-clicking opens it for rewriting.

The pieces can be searched, filtered by state, turned on and off,
deleted,
rewritten and added to, one at a time or in a batch. Every one of those
was
already on the contract with no reader for it. A base's settings are
back and
fuller: how many pieces a search returns, how documents are cut, and how
much
prose comes with a table or a figure. The embedding model is a picker
while the
base holds no pieces and a stated fact once it does -- moving it then is
the
collection made again at the new width and every document re-cut, which
is not
a thing a panel does by being saved.

**An agent can read them.** The retrieval engine was there and nothing
could
reach it: no agent had a tool that opened a library. A reader now points
a
conversation at bases from the composer, the way they pin a working
directory,
and the selection rides in the same place -- the session's metadata,
bound
around the turn.

The tool takes a query and never a base. The scope is the reader's
choice, so a
model cannot reach material nobody offered it, and it is withheld where
nothing
was picked rather than offered and refused. A hit says which document
and page
it came from, because a hit a reader cannot place is a hit they cannot
check --
and the passages are pressable in the transcript, walking into the base,
the
document, the folder it is filed under and then the piece. A citation a
reader
cannot follow is a claim they have to take on trust, and the whole point
of
searching their own material is that they can go and look.

**The word segmenter and the resources.** `raven.core.Tokenizer` is
Infinity's
`rag_tokenizer`, ported: a forward and backward maximum match over a
dictionary
trie, with stemming and lemmatisation on the latin path. XGBoost is not
carried
over. None of the data lives in git -- the dictionary, the NLTK corpora
and the
ONNX graphs run to about a hundred megabytes, so `make fetch-resources`
on an
install and an optional Docker layer behind `RAVEN_RESOURCES` both point
at one
script. A checkout that could not reach the network still works, and
every
feature that reads this data says so when asked.

## Type

- [ ] Fix
- [x] Feature
- [ ] Docs
- [ ] CI / tooling
- [ ] Refactor
- [ ] Other

## Verification

Run on macOS against Python 3.12, with the deepdoc weights fetched (the
model
tests skip without them):

```
uv run pytest                        27450 passed, 111 skipped, 3 failed
npm run test --prefix ui-web          3276 passed
npm run test --prefix ui-tui          2106 passed
npm run type-check --prefix ui-web    clean (tsc)
npm run lint --prefix ui-web          clean (4 pre-existing hook warnings)
make lint-python                      clean (ruff check + format)
make lint-types                       clean (ty)
make lint-imports                     10 contracts kept, 0 broken
make check-source-language            clean
make check-large-files                clean
make build-ui                         built, both boot snapshots match
node ui-web/scripts/check-css.mjs     clean
node ui-web/scripts/check-class-namespace.mjs  clean
```

The three Python failures are not this branch's and were each confirmed
on the
base alone: one asks a host without LibreOffice to refuse a render and
this host
has LibreOffice installed, and two simulation cases fail on an
unmodified
checkout of the base commit.

The RPC contract was regenerated for both consumers and both match it
(`npm run gen --prefix ui-web`, `npm run gen:rpc --prefix ui-tui`, then
`gen:check`). The page's boot snapshots and region goldens were
refreshed where
the composer bar gained a button, and each records that button and
nothing else.

- [x] Relevant tests pass locally
- [x] Relevant lint / type checks pass locally
- [x] User-facing docs or screenshots are updated when needed

## Risk

A PDF's tables, and a Word or PowerPoint file's, are now indexed as HTML
rather
than as header-prefixed lines. Chunk text changes shape for those
documents, so
a base indexed before this and searched after it holds both forms until
its
documents are reindexed; retrieval keeps working either way, because
both forms
are text. The same applies to how much prose a table or a figure carries
with
it: the setting now reaches as far as it says, where before it stopped
at the
nearest paragraph, so a chunk cut before this keeps the narrower context
until
its document is rebuilt. Rolling back is reverting the branch and
reindexing.

The deepdoc models are optional. An install that never ran `make
fetch-resources` reads PDFs geometrically, exactly as before, and says
so when
the models are asked for by name. Nothing here reaches the network at
request
time.

Chunk crops and rendered pages are written at index time and on first
view, both
under the state directory rather than in the repository, and both are
deleted
with their document. A page is immutable for the life of its document
id, so
the route serving them caches; the document route does not. All three
routes
take ids and no path, and the two image routes answer under the
strictest
sandbox in the tree.

What an agent may search is the reader's choice and nothing else. The
tool takes
no base argument, so a model cannot name one, and it is absent from the
schema
entirely for a conversation that named none -- a lane nobody pointed at
a base
is unchanged by this branch. The bases a session carries are stored
beside its
working directory and travel no further.

- [x] Security impact considered
- [x] Backward compatibility considered
- [x] Rollback path is clear for risky changes

## Related Issues

N/A

---------

Co-authored-by: Claude (claude-opus-5) <noreply@anthropic.com>
2026-10-12 19:05:16 +08:00
..

Raven in Docker

Build and run the current checkout:

cd docker
docker compose up --build

To run the configured image without rebuilding, use docker compose up --no-build. Plain docker compose up may reuse or pull an image rather than rebuild local changes.

Then open http://localhost:18793. The local Compose default signs that browser in automatically so Settings > Models is immediately available.

What comes up

One container, two processes:

process what it does
nginx (:80) serves the built page from /app/ui-web/dist, proxies /rpc, /auth*, /health, /file, /files/, /oauth/callback to the engine
the Raven engine the RPC gateway on 127.0.0.1:18793 inside the container

They share a container because the engine binds loopback only (web.TCPSite(runner, "127.0.0.1", port)), so a proxy in a second container would have nothing to dial. The image builds the page in a Node stage and the Python environment with uv sync --frozen, so both halves come from the same commit.

Signing in

The engine mints a session credential per boot and the page needs a cookie for it. Compose defaults RAVEN_AUTO_LOGIN=1, so nginx gives a browser loading the page that credential automatically. This is intended for a locally published port.

For a remotely exposed deployment, put RAVEN_AUTO_LOGIN=0 in .env.local. The entrypoint prints a one-time link on startup:

raven-docker: sign in once with this one-time link (replace localhost if you are remote):

    http://localhost:18793/auth#<nonce>

The cookie it sets lasts 30 days, and survives a restart -- the engine reloads the same cookie from /data rather than minting a new one.

The page's raven web advice is for a host install. In a container, mint another one-time link with:

docker compose exec raven docker-entrypoint.sh signin

Configuring a provider

The engine starts with nothing configured. It says so on the way up:

! no provider is configured yet -- run `raven onboard` for guided setup
  Starting anyway. Add a provider in Settings > Models on the page.

Add one in the page's Settings > Models and the next turn uses it -- the provider reads its credentials per call, so nothing needs restarting.

To have a deployment come up already configured, declare one in .env.local instead:

RAVEN_PROVIDER=anthropic
RAVEN_API_KEY=sk-ant-...

The entrypoint writes that into config on every boot, so where it is set it is the source of truth.

Local providers can be seeded by address without an API key:

RAVEN_PROVIDER=ollama-chat
RAVEN_API_KEY=
RAVEN_API_BASE=http://host.docker.internal:11434

The container always runs raven gateway, including the page, RPC, IM channels, cron, and the proactivity stack. Its resolving provider re-reads credentials on every call, which is what makes a provider added in Settings work on the next turn.

Settings

.env is committed and holds defaults. The service loads the optional, git-ignored .env.local second for container runtime variables such as provider credentials and RAVEN_AUTO_LOGIN. Put every credential in .env.local.

Compose interpolation is separate: RAVEN_IMAGE, RAVEN_WEB_PORT, RAVEN_EXTRAS, RAVEN_PLUGINS, and RAVEN_OFFICE are resolved before the container starts. Setting them only in the service's .env.local does not override the image, published port, or build arguments. From this directory, use shell variables, for example:

RAVEN_IMAGE=raven:local RAVEN_WEB_PORT=18893 docker compose up --build

Or, after creating .env.local, load it explicitly for interpolation too:

docker compose --env-file .env --env-file .env.local up --build

The later file wins; shell environment values override both files.

Data

Everything durable is under /data on the raven-data volume: config.json, the workspace, sessions, logs and the memory store. Removing the container keeps it; docker compose down -v does not.

Running a CLI command

Any argument that is not run or signin is handed to the raven CLI:

docker compose exec raven docker-entrypoint.sh provider test anthropic
docker compose run --rm raven status

Notes and limits

  • Exposure. The default Compose page automatically grants a Raven browser session to anyone who can reach its port, so the published port binds 127.0.0.1 only. A remote deployment must choose its own bind or proxy and set RAVEN_AUTO_LOGIN=0. nginx also rewrites the Origin header so the engine accepts a browser that reached it through a published port. That drops a defence-in-depth layer; the session cookie is SameSite=Strict, so an unauthenticated socket is still refused. Publish this port only where you would have published the engine's own.

  • HTTPS. Not configured here. Terminate TLS in front of this container; the page picks wss:// from location.protocol on its own.

  • The TUI is not in this image. raven tui needs the Node bundle from ui-tui/, which the image does not build.

  • sandbox. The extra is off by default: inside a container the container is the boundary. Turn it on with RAVEN_EXTRAS if you want boxlite as well.

  • LibreOffice is in the image. The page cannot draw a .docx, .xlsx or .pptx, so the gateway converts those to PDF to show them; without it every office file in a knowledge base, and every deck, answers the viewer with "LibreOffice is not installed" -- a hosted image telling its user to install something on it. It is the largest thing here, roughly 500 MB with the CJK fonts it needs to render anything but Latin text. Build with --build-arg RAVEN_OFFICE=0, or RAVEN_OFFICE=0 in the environment Compose reads, for a deployment that will never preview one.

  • The tokenizer dictionary and the deepdoc models are in the image, about 160 MB. They are fetched at build time rather than at first use so the container needs no egress to read a PDF or to segment Chinese -- a deployment without it would otherwise fail the first time somebody uploaded one, long after the build that could have said so. Build with --build-arg RAVEN_RESOURCES=0, or RAVEN_RESOURCES=0 in the environment Compose reads, for a smaller image that does neither; both features report their own absence rather than failing obscurely.

    Before you push this image anywhere, read the "deepdoc models" section of NOTICES.md. One of the four, layout.onnx, stamps AGPL-3.0 into its own metadata under author: Ultralytics, and its metadata contradicts itself about that. Building the image here is your own fetch happening inside a container; pushing it to a registry, or shipping an appliance built from it, is redistribution, and that is where the question becomes yours to answer. RAVEN_RESOURCES=0 is the build that carries none of them.