Developers

One API. The whole document pipeline.

Send documents and data through one API. An agentic workflow parses, extracts, checks and cites them, then hands the results back as an app for your customers or as an API of your own. Nothing to stitch together.

Read the docs Get API access

One API and MCPREST with OpenAPI, and one MCP endpoint
Python SDKsSteps, connectors, parsing and agents in code
Yours or oursOur vector store and storage, or bring your own

Pipeline as code

Design the whole pipeline in code.

Define every step from ingestion to the app your customers open, in Python, JSON or straight over REST. Run it, stream every event and version it like the rest of your code.

covenant_review.py

1from workbricks import Workbricks, steps2from schemas import ComplianceCertificate3 4wb = Workbricks()  # reads WORKBRICKS_API_KEY5 6flow = wb.workflows.create(7    name="covenant-review",8    trigger=steps.on_file_added(source="gmail"),9    jobs=[10        steps.parse(id="parse", ocr=True, tables=True),11        steps.extract(id="fields", after="parse",12                      schema=ComplianceCertificate, cite=True),13        steps.embed(id="index", after="parse",14                    store="workbricks"),  # or your own15        steps.agent(id="check", after="fields",16                    tools=["credit.lookup_covenant"],17                    instructions="Compare the cited ratio with the agreement threshold. Queue breaches for review."),18        steps.publish(id="out", after="check",19                      app="covenant-monitor", api=True),20    ],21)22 23run = flow.run(files=["q3_compliance.pdf"])24for event in run.stream():25    print(event.step, event.status)

covenant-reviewrun_8f2c · running

Totally flexible

Every stage is yours to swap.

Use what Strive runs for you, or plug in your own at any stage. Mix both in the same workflow.

01Ingest

Built inConnectors for drives, inboxes, databases and APIs

YoursYour own connector in Python, or push records over the API

02Parse

Built inLayout, tables and OCR across PDFs, Office files and images

YoursYour parser, dropped in as a custom step

03Chunk

Built inHeading-aware chunks sized to a token budget

YoursYour own chunking strategy

04Embed

Built inHosted or local embedding models

YoursThe embedding model you choose

05Store

Built inStrive vector store, database and file storage

YoursYour vector database and your own bucket

06Reason

Built inAgents with tools, conditions, retries and review

YoursYour code in sandboxed Python or a container

07Deliver

Built inAn app your customers open, under your name

YoursAn API your own product calls

Workflow functions

Every function a workflow needs.

Agents, document processing, retrieval, code, data and delivery. Chain them in any order, from code or on the canvas, and add your own when you need something new.

41 functions

steps.agentAgents and AI
Plans, calls tools and decides, within the limits you set.
in · recordsout · records
Settings
instructionsstring
What the agent should do
toolsstring[]
Connector, MCP and workflow tools it may call
max_stepsint
Upper bound on tool calls
reviewpolicy
When a person approves before it acts
```
1steps.agent(2    id="agent",3    instructions="...",4    tools=[...],5    max_steps=5,6)
```

steps.agentAgents and AI

Plans, calls tools and decides, within the limits you set.

in · recordsout · records

Settings

instructionsstring: What the agent should do
toolsstring[]: Connector, MCP and workflow tools it may call
max_stepsint: Upper bound on tool calls
reviewpolicy: When a person approves before it acts

1steps.agent(2    id="agent",3    instructions="...",4    tools=[...],5    max_steps=5,6)

Python SDKs

The same building blocks we use.

Write your own steps, connectors and MCP servers with the SDKs the platform itself is built on. Package them, register them, and they show up in the workflow builder for your whole team.

$ pip install workbricks-workflow

workbricks-workflowSteps, ports and workflow composition
workbricks-connectorsSources and destinations, with OAuth handled
workbricks-docparseThe document parser, with page-anchored citations
workbricks-agentsAgents, tools and human review
workbricks-mcpStart an MCP server from a template, or from an OpenAPI spec
workbricks-sandboxRun untrusted code in isolated sandboxes

match_purchase_order.pya custom step

1from workbricks.workflow import handler, PortSpec, StepContext2 3@handler(4    name="match_purchase_order",5    version="1.2.0",6    inputs={"invoices": PortSpec(kind="records")},7    outputs={"matched": PortSpec(kind="records")},8    side_effect="none",9)10class MatchPurchaseOrder:11    async def execute(self, ctx: StepContext, invoices: list[dict]) -> list[dict]:12        erp = await ctx.connector("erp")13        out = []14        for inv in invoices:15            po = await erp.get("purchase_orders", inv["po_number"])16            gap = inv["total"] - po["total"]17            out.append({**inv, "po_total": po["total"], "flag": gap > 10_000})18        ctx.log.info("matched", count=len(out))19        return out

Production ready

Built to run, not just to demo.

Everything you need to put a pipeline in front of customers and keep it there.

REST and OpenAPIEvery service publishes an OpenAPI reference. Generate a client in any language.GET /v1/openapi.json
One MCP endpointEvery tool through a single endpoint, each call recorded and changed tools held for review.POST /mcp
Streaming eventsFollow every step of every run as it happens, over server-sent events.Accept: text/event-stream
Retries and timeoutsPer-step retries with backoff, timeouts and an error policy for each step.max_retries: 3
Versioned workflowsEvery change is a version. Restore any one, and runs keep the version they started on.PUT ?expected_version=12
Tracing and metricsOpenTelemetry traces and metrics for every run, step and model call.workflow_step_duration_seconds
Webhooks and schedulesStart runs from a webhook, a schedule, a new file or a new record.trigger: cron 0 6 * * 1-5
Your cloud or oursRun on Strive, or install it in your own Kubernetes cluster with Helm.helm install workbricks

Questions

What engineers ask.

Is our data used to train models?

No. Your documents and results are never used to train a model, by us or by the AI services we use. Our data processing addendum puts that in writing.

Is our data encrypted?

Yes. AES-256 at rest and TLS in transit, and each organization's data is kept apart at the database level. Our security controls are independently audited under SOC 2 Type II.

Is API access included?

Yes, on every agreement. The API calls the same workflow as your app, and runs are unlimited.

What happens when a step fails?

It retries with backoff up to the limit you set, then follows that step's error policy. Point the Assistant at a failed run. It reads the run history, explains the problem and proposes a fix.

Where does custom Python run?

In a sandbox. Your code runs in an isolated container with limits on what it can reach on the network, and with a hard timeout you set.

Can we self-host it?

Yes. Install it in your own Kubernetes cluster with Helm, in your own cloud account or on your own hardware. Self-hosted installations come with metrics, logs, traces, dashboards and alerts.

Can I use my own model API key?

Yes, on any agreement. Model costs then go straight to your account with the vendor. The AI that reads pages and pulls out fields is already included in the page price.

Start building.

Read the docs, get an API key and run your first workflow in minutes.

Read the docs Get API access