Turn Birk models
into running workflows

Compose Birk models into reliable multi-step workflows with parallel processing, retries, approvals, and full observability.

One workflow,
the right model
at each step

Workflows decide where each Birk model belongs: fast classification, tool-using reasoning, deep analysis, image generation, or voice output. The result is not one long prompt, but a production path you can inspect and improve.

Birk-Fast

Classify, route, summarize, and handle quick text or vision analysis.

Birk-Light

Reason over context, call tools, and produce structured workflow steps.

Birk-Heavy

Take over when long context, code, or high-depth analysis is required.

Blink

Generate visual outputs from approved prompts or refined briefs.

Blip

Deliver realtime voice output and conversational audio experiences.

Better way with

Text Agent

Classifies intent, retrieves context, and writes a grounded response.

Birk-FastBirk-Light
1
Understand request
Detect intent, urgency, language, and required policy scope.
2
Fetch trusted context
Pull only permitted CRM notes, policies, and prior tickets.
3
Generate grounded answer
Write the response with citations and no unsupported claims.
4
Validate before send
Check tone, missing evidence, and escalation conditions.
Text Agent
1import { agent, task } from "@briqmind/sdk";2 3const textAgent = agent("support-writer", {4  model: "birk-agent-light-v1",5  tools: [retrieveContext, qualityCheck],6});7 8export const answerCustomer = task(async (message: string) => {9  const intent = await textAgent.classify(message);10  const context = await retrieveContext(intent);11 12  const draft = await textAgent.generate({13    message,14    context,15    style: "concise",16  });17 18  return qualityCheck(draft);19});

Break work into reliable steps

import { task } from "@briqmind/sdk/workflows";

const analyzeDocument = task(
{ name: 'analyzeDocument' },
function analyzeDocument(doc: string): string {
console.log(`Analyzing document with Birk`);
return birk.extract(doc);
}

Define model-sized tasks

Turn a request into clear Birk steps: route with Fast, reason with Light, escalate to Heavy, then call tools or produce output.

#Birk-Fast#Birk-Light#Birk-Heavy

Retry the risky step only

Retry the failed model call, connector, Blink render, or Blip voice segment without repeating the entire workflow.

#tool-call#Blink#Blip

Save context and resume

Store inputs, retrieved context, model outputs, approvals, and tool results so long-running work continues from the right point.

#context#approval#resume

Scale model work without waiting

AVG QUEUE TIME: 230 MS

Fast decisions stay responsive

Keep Birk-Fast classification, routing, and lightweight analysis separate from heavier jobs so everyday decisions do not slow down.

#Birk-Fast#routing
CONCURRENT TASKS: 2232

Model chains run in parallel

Run Birk-Light, Birk-Heavy, Blink, and Blip steps at the same time so one long task does not block the whole workflow.

#Birk-Light#Birk-Heavy#Blink#Blip
Task
> 22hrs
Sub-Agents8
Time8h 40m

Heavy work gets dedicated capacity

Give long-context analysis, large document processing, and generation jobs their own room to finish cleanly.

#Birk-Heavy#long-context

Every model step. Full visibility.

1,402Total runs
12Running
4Waiting
100500
10:00pm11:00pm12:00pm1:00am2:00am
Succeeded
Failed
RUN ID STATUSSTEPSTARTEDDURATION
trn-d52pv8f5r7bs738gcq00
Waiting
policy_context1 hour ago-
trn-f8k3v2m9q7bz4lx6p0rt
Running
tool_call.crm_update1 hour ago32s
trn-a6w9c4r8y1p5z0t3n7hs
Running
birk_light.reason1 hour ago41s
trn-m2q7v8x1k5r9c0f3z6pbSucceededblink.render2 hours ago23s
trn-r5x0t8k3p6v2q1z9b4mcSucceededbirk_heavy.review2 hours ago14s
trn-b7z3f6p2k9x4r1v8m0qcSucceededblip.voice_reply2 hours ago93s
Run ID: run_birk_7f31
Succeeded
Customer request workflow
Birk-Fast routing
Attempt 2
Birk-Light reasoning
Attempt 1
CRM context returned incomplete
Tool call: update CRM
Approval check completed
Response prepared for user
Run details

Tool call: update CRM

Status Succeeded
Duration243 ms
Cost$0.01
Triggered
Attempt 1
Attempt 2
Succeeded

Trace every model call, tool call, approval, retry, latency spike, and generated output in one run history.

Compare Birk models' performance across time, cost, reliability, and failure reason.

Inspect the exact step that needs work: prompt, retrieved context, connector response, model output, or human review.

Pay per use

Simple pricing based on the number of concurrent workflows you need to run and the compute you use, billed by the second. Zero per-task fees.

Explore pricing
Workspace plan
Max concurrency: 5 tasks
Guaranteed warm starts:
Task compute$/secCPURAM
Starter$0.050.5 CPU512 MB
Standard$0.201 CPU2 GB
Pro$0.402 CPU4 GB
Node.js
Python
TypeScript
vLLM
NVIDIA
Docker
VS Code
PyTorch
Kubernetes
Git
Bun
React
Next.js
Vercel

Start a workflow
with Birk

Choose the right model, connect your tools, and move reliable AI workflows into production.

Build a workflow