Unmeshed is a workflow orchestration platform that runs APIs, AI agents, batch jobs, human approvals, and long-running workflows with durability, observability, and governance built in. It was built by the creators of Netflix Conductor.

How is Unmeshed related to Netflix Conductor?

Unmeshed was built by the creators of Netflix Conductor as a next-generation orchestration platform, and it offers a clear migration path for teams running Conductor today.

Can one workflow combine code, AI agents, and human approvals?

Yes. Steps in an Unmeshed workflow can be code, HTTP calls, decision tables, AI or agent calls, and human tasks. A workflow can pause durably for a human approval — for minutes or weeks — and resume exactly where it left off.

How does Unmeshed keep LLM costs under control?

Deterministic steps cost nothing in tokens, so models run only where they add value. You can configure token limits per AI step, control which tools an agent can reach, and attribute cost to each workflow — every execution respects those settings.

Which AI models can I use with Unmeshed?

Unmeshed works with OpenAI, Claude, Gemini, Amazon Bedrock, Mistral, and Ollama, and supports the Model Context Protocol (MCP). You choose the model per step or per agent.

Which languages do the SDKs support?

Unmeshed ships SDKs for Java, Go, TypeScript/JavaScript, and Python, plus a CLI for local runs and CI. Hosted functions also let you run step code in Python, JavaScript/TypeScript, or Go without separate hosting.

How much does Unmeshed cost?

Unmeshed has a free tier you can start on today, plus Premium and Enterprise plans as you grow. See the pricing page at unmeshed.io/pricing for current plans.

From the creators of Netflix Conductor

Code, AI, rules, and people. One workflow.

APIs. AI agents. Batch jobs. Human approvals. Unmeshed turns all of it into one workflow you can watch happen in real time.

Poke around below, no signup required.

Start for Free Book a Demo

Filter by step typePlay a request

Every step declares what it is: code, a model call, a rules table, or a person. One workflow runs them all with one audit trail.

Process

Code, model, rules engine and person in one process, one runtime, one audit trail.

Run record

Every step writes its input, output and decision to one run record, including which person decided. Start a run to watch it fill.

Code step(runs on hosted functions)LLM step(governed model call)Decision table(decision engine)Human task(approval checkpoint)Integration(100+ built-in apps)

Built by creators of

•originally built at

and trusted by

Inside the Run

You just watched these steps run. Here's each one, up close.

The demo above pauses for a person, checks a rules table, and runs code with zero infrastructure. Three shipping features, up close.

Human in the loop

You already did this; the approve button in the demo was a real human-in-the-loop step. Workflows pause durably for approvals, for hours or weeks, then resume exactly where they left off.

API Call via GET

# get_processing_data

Wait for Human Input

# wait_human_review

Flow Gateway

# check_input

CONTINUE

API Call via POST

# post_processed_data

See human in the loop

Decision Engine

The approval routing you watched came from a decision table, business logic your whole team can read, versioned, testable, and deterministic every time.

See the Decision Engine

Hosted functions

The code steps ran on hosted functions. Write a step in Python, JavaScript/TypeScript, or Go; Unmeshed runs it. No containers to build, no separate hosting, no deploy pipeline.

See hosted functions

normalize_input.pyhosted function

def normalize(expense):
  has_fields = "amount" in expense \
    and "category" in expense
  return {"structured": has_fields}

step completed — no server to deploy

The Platform

Three products. One orchestration engine.

A multi-step workflow, a customer-facing API, a nightly batch run. Different shapes, same durable core underneath. Every step is checkpointed, retried, and audited by the same engine.

Product · Workflows

Automate multi-step processes that survive anything.

Kill the pod, deploy on a Friday, wait ninety days for a signature, the workflow picks up exactly where it left off. Retries, versioning, branching, and full run history are built in, not bolted on.

Explore Workflows

Example Flow

Support ticket triage

ingest

ticket received

crm_lookup

api · crm

kb_search

kb · context

classify

llm · triage

route

rules · table

review

human · review

assign

route · assign

Deterministic step · $0AI step · billed only hereHuman-in-the-loop · first-class

Mobile Apps

Devices

Chats

Emails

Webhooks

Web Apps

Databases

Internal APIs

External APIs

SaaS Apps

Queues & Streams

Other Sources

Product · API Orchestration

Serve workflows as high-scale API endpoints.

Expose any workflow as a synchronous API. Fan out to your services, aggregate the results, and respond in milliseconds, with the engine handling retries and fallbacks behind the endpoint.

Explore API Orchestration

Product · Batch Job Processing

Batch scheduling at second-level granularity.

Cron gives you a start time. Unmeshed gives you the whole run, second-level schedules, managed dependencies between tasks, and a live view of every job in flight.

Explore Batch Processing

05:00

05:01

05:02

05:03

warehouse_nightly_load

05:04

05:05

partner_file_ingest

05:06

05:07

invoice_settlement_post

05:08

05:09

eod_reconciliation_run

05:10

05:11

compliance_report_gen

05:12

05:13

payroll_batch_close

05:14

05:15

warehouse_nightly_load

05:16

05:17

partner_file_ingest

05:18

05:19

invoice_settlement_post

05:20

05:21

eod_reconciliation_run

05:22

05:23

compliance_report_gen

05:24

05:25

payroll_batch_close

05:26

05:27

warehouse_nightly_load

05:28

05:29

partner_file_ingest

05:00

05:01

05:02

05:03

warehouse_nightly_load

05:04

05:05

partner_file_ingest

05:06

05:07

invoice_settlement_post

05:08

05:09

eod_reconciliation_run

05:10

05:11

compliance_report_gen

05:12

05:13

payroll_batch_close

05:14

05:15

warehouse_nightly_load

05:16

05:17

partner_file_ingest

05:18

05:19

invoice_settlement_post

05:20

05:21

eod_reconciliation_run

05:22

05:23

compliance_report_gen

05:24

05:25

payroll_batch_close

05:26

05:27

warehouse_nightly_load

05:28

05:29

partner_file_ingest

Agentic AI

AI agents that are governed and production-ready.

Intelligence is table stakes. Reliable execution is the differentiator. Unmeshed agents use tools that invoke real workflows and governed endpoints, with guardrails and full run observability.

Model router

governed

AI Output

“High intent B2B lead, score 87 — route to enterprise queue and notify AE.”

Structured output

{"intent": "high","score": 87,"route": "enterprise_queue"}

calls workflow: score_lead

Live · workflow step

The agent builder, as it ships today, click to inspect.

Build a governed agent

Cost Control

Spend tokens only where they earn their keep.

Deterministic steps cost $0 in tokens. Set token limits per AI step and every execution respects them, no runaway LLM spend. See which steps actually bill, then flip the toggle in the comparison to price the two architectures.

Workflow stepsExecution type

parse_input

crm_lookup

classify_intent

route_decision

assign_agent

4 code steps

$0.000 each

1 llm step

$0.018 billed

cost / workflow

$0.018 vs $0.090

parse_inputcode · $0.000

crm_lookupcode · $0.000

classify_intentllm · $0.018

route_decisioncode · $0.000

assign_agentcode · $0.000

Deterministic step · $0LLM step · only where needed

Start from a workflow template

Illustrative per-step costs, your numbers depend on your workflows.

Step 01

parse_order

code $0.000

Step 02

validate_totals

code $0.000

Step 03

lookup_customer

code $0.000

Step 04

classify_risk

llm $0.027

Step 05

route_decision

code $0.000

Step 06

enrich_address

code $0.000

Step 07

check_inventory

code $0.000

Step 08

choose_carrier

code $0.000

Step 09

write_audit

llm $0.008

Step 10

commit_order

code $0.000

Cost · This Run

$0.035

2 llm steps

Steps Billed

2 / 10

Projected · 500k Runs/Mo

$210k

annual

See full pricing breakdown

Illustrative example, your costs depend on your workflows.

Convinced? Run a workflow like the one you just watched, free.

Start for Free

Integrations

Your stack is already connected.

Built-in connections to over 100 apps, plus first-class HTTP, events, and webhooks. The orbit rotates through real integrations we ship today, no placeholders.

Protocol steps (HTTP, GraphQL, gRPC, webhooks, events)
AI providers (OpenAI, Claude, Gemini, Bedrock, Mistral, Ollama, plus MCP)
Databases and warehouses (Postgres, MySQL, BigQuery, and more)
Messaging and queues (Kafka, SQS, Slack, email)
Monitoring, CRMs, and the SaaS tools your teams already use

Browse all integrations

API to Worker Execution Path

Fetch customer order + inventory context

GET /orders/{id} + GET /inventory/reservations

API Step

Run allocation worker in application code

worker.allocateOrder(order, inventorySnapshot)

Worker Step

Persist allocation decision downstream

POST /fulfillment/allocations

API Step

Fetch customer order + inventory context

GET /orders/{id} + GET /inventory/reservations

API Step

Run allocation worker in application code

worker.allocateOrder(order, inventorySnapshot)

Worker Step

Persist allocation decision downstream

POST /fulfillment/allocations

API Step

Developer Experience

SDKs for Java, Go, TypeScript, Python, and C#.

Write workers anywhere, the engine handles scheduling, retries, and state. A CLI covers local runs and CI, and hosted functions run step code with no infrastructure at all.

Explore the SDKs CLI tooling

Production Ops

Built for the runs you can't afford to lose.

Second-by-second observability, enterprise-grade security, and throughput engineered for scale. This is the ops floor your workflows run on.

0

steps per second

Per sharded instance, engineered for production scale.

Run history and metrics on every execution, click to inspect.

Enterprise-grade security

SSO and SAML login, SCIM provisioning, role-based access control, and audit logs on every action across workflows and endpoints.

Read about security

Observability on every run

Full run history, per-step metrics, and searchable executions, so even your AI steps stay visible and manageable.

See deep insights

Durable by design

Pause, resume, and retry without losing state. Workflows survive restarts, deploys, and month-long waits.

Compare durable execution

FAQ

Questions, answered.

Untangle your workflows.

Start free and ship a durable workflow this afternoon, or book a demo with the team that built Netflix Conductor.

Start for Free Let's talk

Code, AI, rules, and people. One workflow.

You just watched these steps run. Here's each one, up close.

Human in the loop

Decision Engine

Hosted functions

Three products. One orchestration engine.

Automate multi-step processes that survive anything.

Serve workflows as high-scale API endpoints.

Batch scheduling at second-level granularity.

AI agents that are governed and production-ready.

Spend tokens only where they earn their keep.

Your stack is already connected.

SDKs for Java, Go, TypeScript, Python, and C#.

Built for the runs you can't afford to lose.

Enterprise-grade security

Observability on every run

Durable by design

Questions, answered.

What is Unmeshed?

How is Unmeshed related to Netflix Conductor?

Can one workflow combine code, AI agents, and human approvals?

How does Unmeshed keep LLM costs under control?

Which AI models can I use with Unmeshed?

Which languages do the SDKs support?

How much does Unmeshed cost?

Untangle your workflows.