Data layer
One operational source of truth every agent can use.
ERP, CRM, TMS, spreadsheets, and field systems are unified into typed workspace data, so agents execute from live context instead of stale exports.
Under the Hood
The engine that powers your operation.
Five layers — Data Hub, Watches, Agents, Feeds, and Governance — assembled and operated by us. This is the infrastructure behind every briefing, alert, and scorecard we deliver to your team.
5runtime layers
9workspace surfaces
12feed types
48hrto first briefing
What runs underneath
The operating spine that keeps your workspace live between reviews.
Multiple workspace surfaces share one foundation: live data, continuous watches, generated intelligence, governed tools, and a memory of what happened before — so the operation keeps working without waiting for someone to refresh it.
Watch engine
Every condition can be monitored continuously.
Thresholds, record changes, SLA risks, and exceptions run through always-on watches, giving agents the signal they need before the next review cycle.
Intelligence studio
Briefings, updates, and summaries are generated from live data.
The platform turns current records into morning briefings, risk summaries, customer updates, collections lists, and other feed outputs on cadence.
Actions
Agents can move work forward under governance.
Tool grants, scripts, approval tiers, and audit trails let agents route escalations, draft communications, update records, and pause for approval when consequences matter.
Learning loop
The operating model gets sharper as the workspace runs.
Runtime events, knowledge items, and context snapshots help agents remember what worked, which risks preceded outcomes, and where new automation would help.
Workspaces
Your project needs one home, not another loose stack.
The workspace houses the brief, schema, data, analytics, automations, feeds, groups, tools, agents, tasks, and approvals for your project. It can start as reporting, become automation, and later support agents without moving to a new platform.
A unified project data layer
Every workspace runs on a typed schema — entities, fields, relations, attachments, and records with real columns. Analytics, automations, feeds, and agents all work from the same data.
Monitoring rules that become outputs
Watches evaluate thresholds, segments, KPIs, and record-level changes. When a condition fires, it can create a feed, trigger an automation, or invoke an agent/tool path.
Analytics as a native capability
The SQL workbench, KPI model, saved reports, and feed publishing flow let the workspace answer questions and deliver intelligence from the same schema.
Workspace Runtime
One governed project environment
The platform links data, automation, generated outputs, callable tools, schedules, and approval policy so each project is delivered as one operated workspace.
Audience-specific outputs
Twelve feed types — briefings, scorecards, alerts, statements, customer updates, documents — generated from workspace data and shaped for each group.
A governed action surface
Native tools, Python, webhooks, and connected integrations let automations and agents act without hard-coding every project from scratch.
Structural governance
Approval policy travels with the workspace. Low-risk work can run or draft automatically; consequential work queues for human review.
The Runtime
Five pieces. One runtime. One source of control.
Data Hub, Watches, Agents & Automations, Feeds, and Governance — the same five pieces named on the homepage, here in depth. Each layer keeps the operational picture accurate, current, and actionable. Built and proven in manufacturing first.
Layer 1 — Integrate
Data Hub
Entity schemas, connectors, and imports — including inbound documents from suppliers and partners, extracted automatically without manual data entry — into one live operational model that agents, analytics, and humans trust.
- Typed entity schemas — customers, orders, jobs, inventory, costs
- Connectors pull from ERP, spreadsheets, and APIs
- Inbound documents (supplier confirmations, delivery notes, inspection reports) extracted automatically into typed records
Layer 2 — Analyse
Watches
Analytics, KPI computation, passive proposals, and Analyst Watches scan live records continuously. Exceptions surface before the daily stand-up, not after. The analyst layer proposes new feeds, watches, and automations based on detected patterns.
- SQL workbench with NL→SQL and full schema context
- KPI definitions on typed tables
- Watches fire feeds or automations when conditions match
Layer 3 — Execute
Agents & Automations
Agents run as named roles with explicit triggers, tool grants, action capabilities, and approval policies. Automations provide the script and tool substrate for repeatable work. Every agent action is scoped by a deny-by-default tool allowlist.
- Named agent programs with domain triggers and tool policies
- Deny-by-default tool grants per agent
- Python automations and 250+ integrations via MCP and Composio
Layer 4 — Output
Feeds
Feed Studio generates 12 feed types from live workspace data via two authoring paths: LLM generation for briefings and alerts where the analyst voice matters, and uploaded Jinja templates for compliance packs, certificates, and branded quotes that must be identical every time. Feeds are versioned artifacts delivered to groups on a committed schedule.
- 12 feed types — briefings, alerts, customer updates, documents
- Two authoring paths: AI-generated for prose deliverables; Jinja templates for structured documents that must be identical every time
- Briefing arrives before the meeting that needs it
Layer 5 — Govern
Governance
Every action an agent can take is assigned a governance tier based on how consequential it is. Low-risk actions run automatically. Customer-facing or financially material actions require explicit approval. Every run is logged with reasoning, action, and who approved it.
- Four approval tiers from silent to admin-only
- Tier inherited from tool side-effect class
- Full audit trail on every agent action
Setup Spine
From project brief to configured workspace.
Describe your operation. Zipdata proposes the entity model, KPI definitions, automations, feed templates, delivery groups, and agent configurations needed for that project type. Each step is reviewed before it becomes part of the workspace.
Business Brief
Plain-language description of the operation. Zipdata generates entity models, KPI definitions, and connector scripts from this brief.
Entities
Entity schema is proposed and approved step by step. Every field gets a typed column — text, number, date, select, relation.
Connectors
Connector scripts are generated for every source system. Pull schedules, authentication, and field mappings are configured here.
Mock Data
E2B sandbox runs Python scripts to seed realistic records. The workspace is operational before live data is connected.
Metrics & KPIs
KPI SQL formulas are proposed from the entity model. OTIF, SLA, margin, and custom metrics are defined and verified.
Analytics
Initial SQL queries are generated. The workspace analytics library starts populated with the KRAs that matter.
Watches
Analyst Watches are configured: thresholds, segment conditions, and trigger modes. Each watch has an explicit output action.
Feeds
Feed Studio builds the first briefings, scorecards, and alert templates from live query data. Delivery schedule is committed.
Publish
Groups are created, delivery is activated, and the workspace goes live. First automated briefing within 48 hours.
Outputs & Actions
Outputs and actions stay connected to the project.
Feeds create the client-facing outputs. Groups personalize who receives them. Tools let automations and agents perform controlled actions. Together, they turn the workspace from a data project into an operated delivery environment.
NativePlatform actions
Core reads, writes, KPI computation, watch evaluation, feed generation, and entity write-backs built directly into the platform. Zero latency, no auth overhead.
→ Execute analytics query→ Write entity record→ Trigger feed generation
PythonPython sandbox
Sandboxed Python for transforms, forecasting, enrichment, and file processing. Secrets are injected; the sandbox has no default network access.
→ Run forecast model→ Transform CSV→ Enrich records from external API
WebhookHTTP integrations
Outbound HTTP contracts for ERP write-backs, legacy systems, and custom notification pipelines. Payload schema is defined on the tool.
→ Update ERP order status→ Notify Slack channel→ Push to legacy system
MCPSelf-hosted MCP
Model Context Protocol integrations for regulated environments where data must stay within your trust boundary. Bring your own MCP server URL.
→ On-premise ERP write→ Internal CRM action→ Proprietary API call
250+ appsComposio integrations
Managed OAuth access to Salesforce, Jira, Gmail, HubSpot, Slack, Linear, ServiceNow, and 245+ more. OAuth connections are managed per workspace.
→ Create Jira ticket→ Send Gmail draft→ Update Salesforce record
TestMock / sandbox
Schema-generated synthetic responses for demos, agent development, and approval-flow testing — no live side effects.
→ Test agent flow→ Demo approval UI→ Validate tool schema
Governance Model
Every action has a tier. Every tier has a policy.
Approval tiers derive from each tool's side-effect class — not from per-agent configuration. Add a new tool and it inherits the right tier automatically. The governance surface does not expand as your operation grows.
SilentLog a metric snapshot, tag a record, update an internal counter.Executes immediately. No record in Pending Actions. Audit trail only.
Auto draftGenerate a weekly briefing, draft an internal scorecard.Executes immediately and is saved as draft. Reviewable but not blocked.
NotifySend an internal Slack message, create a Jira ticket.Queues as pending. Human approves before execution. Visible to all operators.
Require approvalUpdate ERP status, send a customer-facing email, trigger a payment.Queued as pending. Admin-only approval gate. Dismissed if not approved.
Configure it for your operation
Your workspace, configured. Your first briefing, delivered.
Pick a domain template, seed realistic data, and see your first agent-generated briefing before the demo call begins.