Paperless
Paperless is Kilimanjaro's AI-enabled team workspace extension. Teams chat with AI agents and colleagues in shared workspaces, run AI-generated workflows organised by category, capture rich operational data through native mobile apps, and connect the workspace to remote databases, analytics, and CRMs through the standard extension framework. Built on Kilimanjaro Stack and shipped to the App Store by Formelo, Inc., Paperless is the canonical Trailblazer-class example of an agentic workspace extension — every capability customers experience is composable in their own extensions through the same framework.
Paperless is built on the same extension framework available to every extension class — Anchor and Circuit extensions ship against the same contract.
1. What Paperless covers
| Surface | Role |
|---|---|
| Workspace | Each workspace is a Kilimanjaro project — its own identity boundary, members, data, policies, custody mode, and audit ledger. Users switch between workspaces in the app (e.g., "Bankberry Field Operations" / "Citywatch: Safer Streets"); the platform isolates data per project as a structural property |
| Chat (humans + agents) | Conversation surface where team members exchange messages and engage AI agents. Agents act under the invoking user's identity scope — every action an agent takes is attributed to the user who invoked it and policy-enforced against the workspace's RLS bindings |
| AI-generated workflows | Structured operational tasks organised by category (Facility Management / Transportation / Safety and Security in the demo workspace; team workspaces author their own taxonomy). Workflows compose forms, capture steps, and approval flows; AI agents can draft workflows from natural-language prompts |
| Rich data capture | Native iOS and iPad capture surfaces with offline-first storage — photos, videos, audio, signatures, GPS coordinates with precise location tagging. Captures sync to the workspace's data layer when connectivity returns |
| Integration substrate | Workspaces connect to remote databases (via Data Bridge extensions), CRMs (Anchor-class extensions), and analytics surfaces through the standard extension framework |
Paperless is a full-surface extension: Console UI for workspace and workflow authoring, runtime API for chat and agent invocation, and background-logic functions for workflow orchestration and capture sync. The extension manifest declares each surface against the framework spec.
2. The workspace concept
A Paperless workspace maps one-to-one to a Kilimanjaro project. Each workspace has its own member roster, data collections, custody mode, policy bundle, and audit ledger. Users can be members of multiple workspaces — the mobile app's workspace switcher moves between them — and each workspace's data is structurally isolated from every other workspace by the platform's project-level isolation primitives.
A new user starts in a default demo workspace; team-managed workspaces come online when the owning organisation provisions a Kilimanjaro project and authorises Paperless on it. The organisation administers the workspace through the standard Console — adding members, configuring policies, setting custody mode, and routing high-stakes workflows through Maker-Checker approvals.
3. Chat with agents and colleagues
The chat surface blends human messages and AI-agent participation. A team member posts a message, mentions a colleague, or invokes an agent (@agent patterns) to do work — the agent reads workspace data, calls extension APIs, drafts content, or kicks off workflows.
Every agent action runs through the same gateway calls a human user would make. The agent's effective identity is the invoking user's identity — row-level security still applies, policy still gates, the audit log still attributes. An agent cannot read data the invoking user could not read, cannot write data the user could not write, and cannot escalate beyond the user's role.
This is the structural answer to "what does the agent have access to" — the workspace's policy substrate. The same Unified Policy bundles that govern human reads and writes govern agent reads and writes; the same audit ledger that attributes human actions attributes agent actions, with the additional provenance of which user invoked the agent.
4. AI-generated workflows
Workflows are the operational shape Paperless captures work in. Each workflow is a structured task — a form to fill, a capture sequence to walk through, an approval flow to route, or a composite of all three. Workflows are organised by category for discoverability; the demo workspace ships categories such as Facility Management, Transportation, and Safety and Security, and team workspaces author their own taxonomy.
Workflows can be authored directly through Studio mode in the Console or generated by AI agents from natural-language prompts ("create a workflow for incident reports that captures location and a photo, requires a manager approval, and routes to the security team"). The agent drafts; the human reviews and publishes; the resulting workflow lands as a Paperless workflow record with a version and a policy binding. Subsequent versions roll forward; in-flight captures continue against the version they started under until they sync.
5. Rich data capture
Captures land as structured records in the workspace's data layer. Supported capture types include:
- Photos and video with device-side metadata
- Audio recordings with timestamps
- Signatures (touch capture; preserved as image)
- GPS coordinates with precise location tagging and on-screen map preview
- Free-form fields (text, number, single-choice, multi-choice, date / datetime, dependent dropdown sourced from a workspace collection)
- File attachments beyond the capture types above
Captures are offline-first on the mobile surface: the app stores pending captures locally and syncs when connectivity returns. The app caches workflow definitions, dropdown-source datasets, and pending captures across app restarts and OS reboots; sync is automatic on connectivity restoration and manual via pull-to-refresh.
Conflict resolution is server-authoritative: if a shared-dataset value used at capture time has changed since the device last refreshed, the submission still validates against the captured value but flags the discrepancy in meta.warnings on the resulting record.
6. Submission provenance — the _capture column
Paperless preserves capture provenance by writing a reserved JSON column _capture on every record it inserts. The column is gateway-injected at write time and carries the workflow version, capture surface (mobile / chat-uploaded / console-edited / widget), actor identity, captured-at and synced-at timestamps, and file references.
_capture.capturedAt is the device-side timestamp for offline mobile captures and the server-side receive timestamp for chat-uploaded, widget, and Console captures. _capture.syncedAt is always the server-side moment the record was written.
The reserved-column convention applies platform-wide: customer schema migrations cannot declare _-prefixed columns (the gateway rejects with RESERVED_COLUMN_NAME_PROHIBITED); customer SDK insert and update payloads cannot include _-prefixed keys (rejected with RESERVED_COLUMN_WRITE_PROHIBITED); only extension write-paths (the Paperless BFF and its background-logic functions) are permitted to write _capture. This makes the provenance record tamper-resistant from the customer application layer.
Customer reads of _capture are unrestricted; the column is present on every Paperless-inserted record and is filterable through the standard data plane API.
7. Capture surfaces
7.1 Paperless mobile app
The Paperless mobile app — iOS and iPad — is the canonical capture surface. Users log into a workspace, browse the available workflows organised by category, and complete captures with offline-first storage. The app communicates with the platform via the Paperless BFF (Backend-for-Frontend) — a set of dynamic routes on the API class at <orgSlug>.api.k9o.io/@<projectSlug>/extensions/paperless/bff/... — not a browser iframe. BFF responses carry a top-level version attribute for forward and backward compatibility across app versions in production.
7.2 Embeddable web widget
For sharing a workflow on a public website (job applications, customer-feedback intake, etc.), Paperless exposes an iframe-mountable widget served from the platform's user-content host (https://<label>-cdn-<rgn>.<usercontent-domain>/extensions/paperless/<version>/iframe/widget/<workflow-id>). The widget honours per-workflow anonymous-submission policy; whether unauthenticated submissions are accepted is a workflow-level setting bound at publish time.
7.3 In-app chat capture
A capture can begin inside a chat conversation — a user or agent attaches a photo, voice note, or signature directly; the attachment lands as a capture record bound to the active workflow or as a standalone capture in the workspace's media collection.
8. Storage and querying
Captures from every surface land in the workspace's own target collection — the collection selected when the workflow is published. The customer collection is the single source of truth; Paperless does not maintain a parallel ledger.
Records are queryable through the standard data plane API — Paperless does not expose a parallel query surface. The same Unified Policy bundles that govern other collection data govern Paperless capture records. The _capture column is filterable alongside any other field in the collection:
// Fetch all mobile-captured records from the last 7 days
await k9o.from('site-audits')
.select()
.filter('_capture->>captureSurface', 'eq', 'mobile')
.gte('_capture->>capturedAt', sevenDaysAgo.toISOString());
Uninstalling Paperless does not delete capture records. The workspace's target collections and their data remain intact after uninstall; the _capture column persists as an ordinary JSON column on existing records. Only Paperless' own extension schema (workflow registry, agent-configuration metadata) is dropped on purge — workspace-owned data is never touched.
9. Permissions Paperless requests at enable
| Permission | Why |
|---|---|
documents/read,write/* | Writing capture records to any workspace-selected target collection |
collections/read/* | Reading shared datasets for dropdown fields; resolving target collection schemas at workflow-authoring time |
files/read,write/<paperless-attachment-bucket> | Storing attached photos, videos, audio, and signatures |
users/read/* | Resolving operator identity for capture attribution and agent-invocation chain |
webhooks/subscribe/* | Firing capture and workflow events for downstream functions and integrations |
The grant binds at enable time. A PROJECT_ADMIN reviews each line before authorising. Agent-invocation permissions are scoped to the invoking user's identity at evaluation time; the agent cannot exceed the user's authority.
10. Events
Paperless emits events through the platform webhook surface — same delivery mechanics as SIEM forwarding: HMAC-SHA256 signature, exponential-backoff retry, at-least-once delivery, 48h pause-and-resume on persistent endpoint failure.
Events fire for workflow publication and retirement, capture creation and update, chat-message delivery, and agent-action attribution. Subscribe to events by binding a function to the relevant trigger or by subscribing a webhook via the project's webhook surface.
11. Integration substrate
Paperless workspaces are typically composed with other Kilimanjaro extensions:
- Data Bridge extensions — bidirectional sync between a Paperless workspace's data layer and a legacy system of record (Oracle, MS SQL Server, MongoDB, etc.) — useful when Paperless is the field-capture surface and a separate database remains the system of truth
- Context — index Paperless captures for semantic search across submission records, files, and chat history
- Lookout — visualise Paperless capture and workflow records with dashboards composed by category
- Gate Studio — author branded login pages and Maker-Checker approval flows for high-stakes Paperless workflows
12. See also
- Extensions API reference — the framework contract Paperless installs against.
- Build Your Own Extension — author your own extension following Paperless' shape.
- Guides → Unified Policy — the policy substrate governing workspace data access (agent and human alike).
- Guides → Maker-Checker — multi-signature approval flows that compose into Paperless workflows.
- API → Data — query Paperless captures through the standard data plane.
- API → Compute — wire functions to react to Paperless events.
- Context — semantic search across Paperless content.