The summer release: CloudRaker, the public API, and playbooks that work with people
This is a big one — we hope you’re as excited as we are!
RakerOne is now simply “CloudRaker”
Same product, simpler name. The app now lives at
app.cloudraker.com, the API at
api.cloudraker.com, and these docs at
docs.cloudraker.com. Nothing you rely on
changes: existing links, API keys, and integrations keep working exactly as
before.
Introducing the Paperwork API: Paperwork automation from your own code
The new /v1 API puts the whole platform behind a clean, documented
developer surface. One call uploads a document and returns structured data
with citations pointing at the exact page and position every value came
from — no setup inside the app required:
The full surface covers everything the app can do: extract structured data, parse documents to clean text, redact personal information from PDFs and audio, fill form PDFs from your own data, sign documents, chain steps into pipelines, and keep files in a workspace. Don’t have a schema? Describe what you want in plain language and one is inferred for you. Batch endpoints handle many documents in one request, and per-organization rate limits keep usage predictable.
Results can be pushed to you too: register a webhook endpoint and every
delivery arrives signed, with a public key endpoint to verify it came from
us. And official client libraries — @cloudraker/api on npm and
cloudraker on PyPI — give you typed access to all of it, with
ready-to-paste TypeScript and Python snippets on every endpoint in the
API reference.
Paperwork Agents: multi-step automation over the API
The API can also run a whole agent — a reviewed, versioned automation
with several steps, the saved actions it is allowed to take, and the points
where a person has to sign off. Six routes cover the lifecycle: list agents,
read one, start a run, poll it, answer a sign-off, and complete a human
step. Runs are long-lived by design — one blocked on a person waits days,
its input files are never purged on a deadline, and paused is a resumable
state rather than a failure. Five webhook event types
(agent_run.waiting, agent_run.approval_requested, agent_run.task_ready,
agent_run.completed, agent_run.failed) ride the same signed envelope as
every other delivery. See Agents and
Agent runs.
Playbooks learn to work with people
Playbook runs used to be agent-only: if the automation needed a decision, it simply stopped. People are now first-class participants:
- Approvals no longer freeze the run. Sign-off requests go into an approval queue and the run keeps working on everything else. Open the queue, see the file names involved, and decide several at once.
- Any step can be assigned to a person. In the designer, choose whether each task is done by the agent or by a human, and who. Assigned tasks show up on the run page as cards a person can claim, complete, or skip.
- A “My work” inbox. Everything waiting on you — approvals, assigned tasks — lives in one place, with email notifications and per-user notification preferences.
Run reliability got a full pass too: a paused run is a resting state, not a failure, and resuming always settles cleanly — long-running, people-paced work is now the normal case.
The playbook editor itself is smoother: activation tells you exactly what’s missing instead of showing errors by default, activating and archiving ask for confirmation, archived playbooks leave the list, selecting a playbook no longer accidentally runs it, and each step can carry richer instructions — extra context about your process or company that the agent uses when it runs.
[Preview] Trigger Center: let the outside world start your work
This feature is limited to customers on the preview channel
Work doesn’t always start with someone clicking a button. The new Trigger Center gives your organization inbound channels that kick off automation on their own:
- Email channels — each channel gets a stable email address; forward a message there and its attachments flow straight into your workspace. Attachments up to 50 MB are accepted and don’t count against your storage quota.
- Webhook channels — a stable URL your other systems can call. Every channel has its own signing secret (rotatable at any time), archived channels stop accepting traffic immediately, and you can send a test event through the real delivery path to verify your setup before any real traffic depends on it.
Each channel shows a per-delivery history, and channels route directly to a target space and a compatible playbook — an email lands, a playbook runs.
Actions, reorganized around what you’re trying to do
The Actions area is now split into a Library of actions you can add and
a clear view of the ones your organization has installed, grouped by
capability. Picking an action at run time is faster: the picker is
searchable, sensibly grouped, and remembers what you used recently, and
running an action walks you through a short guided stepper instead of one
dense dialog. The configuration page was rebuilt with prefilled defaults,
inline validation, and a guard against losing unsaved edits. Every installed
action also has a short, human-readable name you can edit —
redact-client-files instead of a long id — and the name works everywhere
the id does, including the API.
Review extracted data anywhere — no login required
The grounding review workspace — where a reviewer checks each extracted value against the highlighted spot in the source document — can now be embedded outside the app. Send someone a link and they can review citations, correct values, and save changes without an account, whether the source is a PDF, a transcript, or an audio file. Access is short-lived and scoped to just that review.
A real code editor, everywhere JSON lives
Every JSON surface in the app — extraction schemas, action configurations — is now backed by a proper code editor with syntax highlighting, bracket matching, and inline validation, and every read-only JSON view is highlighted to match. Schema authoring gained a shared field type vocabulary: the same thirteen types you know from data objects — Text, Currency, Date, Select, Email, and friends — are available when building extraction schemas, in the app and through the API. See Field types.
Documents read better, and faster
The default text-recognition engine was upgraded: scanned and photographed documents come out noticeably more accurate, and automatic recognition is back on for image uploads. Audio transcription moved to a faster engine.
And when you hand the platform a file by URL — through the API or a pipeline — it’s processed while it downloads instead of after, so parsing and transcription of linked files start sooner and finish sooner, especially for large audio and video.
Generated files stay connected to their source
When an action produces a new file — a redacted PDF, a filled form, a signed document — it stays linked to the document it came from. The Files tab groups outputs under their source with a Generated badge, and generated files skip re-processing, so your workspace stays fast and uncluttered.
Accountability and control
- Audit trail — every meaningful action in your organization is recorded in an audit log administrators can review from Organization Settings.
- API keys, scoped for automation — a new admin screen manages keys, and a key can be marked API-only so an integration key can never be used to browse the rest of your workspace.
- Plan & Usage — a new settings page shows your plan, credit balance, and a running ledger of what consumed credits.
- Developer mode — a per-user toggle that reveals raw JSON and file-output views for people who want to see under the hood.
A rounder, calmer interface
A polish pass driven directly by our users feedback: a clearer sidebar, reworked project tabs, breadcrumbs that only look clickable when they are, consistent badges and empty states, and Organization Settings screens that match the rest of the app in both light and dark mode.