AI workflows on a visual canvas, executed by Go and Temporal. Every run is durable, crash-proof, and exactly-once — even when your infrastructure isn't.
🇮🇳
A new era of workflow automation — built in India, for the world.
console.orcflows.com/workflows/slack-agent
451 req/ssustained, real workflows
9 mstrigger latency, p50
0 %failures across 216K requests
19 MBheap at 200 concurrent users
How it works
From idea to production in three steps
No orchestration code, no YAML, no infrastructure to babysit. Durability, retries, and state are handled by the engine.
01
Design on the canvas
Drag nodes onto a visual canvas — or describe the workflow in plain English and let the AI generator wire it for you.
02
Connect your stack
120+ typed, authenticated connectors. Slack, GitHub, Stripe, databases, LLMs — plus webhooks and polling for everything else.
03
Deploy and trust it
Temporal tracks every step. Crashes, deploys, network blips — your workflow resumes exactly where it left off.
The problem
Automation tools ship features. They don't ship trust.
We read what actually frustrates people running n8n, Zapier, and Make in production — then built the fixes in, not as an Enterprise upsell.
No real version control
Every save is a version, every promotion is reversible
n8n gates real Git integration behind Enterprise, and even then it can't diff or merge — "pulling overwrites everything." OrcFlows snapshots every save automatically, diffs any two versions inline, and gates dev → staging → prod behind an approval step, built in for everyone.
diff · promote · approve
AI agents are a black box
Every tool call, every reasoning step, every retry
Most agent platforms give you pass/fail. OrcFlows captures the full decision trace per run — tool calls, timing, reasoning — and groups every turn of a conversation into one session timeline, so "why did the agent do that" has an answer.
session timeline
No safe way to test before shipping
Dry-run a change with zero real-world side effects
Simulate mode replays real trigger data through real AI reasoning and real branching logic — but every connector, webhook, and approval is stubbed or auto-resolved. Validate a change without a stray Slack message or a real charge.
real logic, zero side effects
Workflows live only in a database
Push and pull workflows to a Git repo you control
Technical teams shouldn't have to choose between a visual builder and PR review. Push any workflow's definition to your own repo, pull it back, and review automation changes exactly like you review code.
PR-reviewable
Durable execution
Write workflows as if failure doesn't exist
Distributed systems break. APIs time out. Servers die at 3am. OrcFlows checkpoints full workflow state at every step, so none of that is your problem anymore.
Crash mid-step — execution resumes from the last checkpoint, automatically
Exactly-once semantics: no duplicate emails, charges, or API calls
Human approval gates that wait indefinitely without timing out
Full event history and replay for every single run
execution · order-fulfillmentrunning
✓
charge_card142ms
stripe.charges.create
✗
worker crashed
SIGKILL — mid-execution, before step 2
↺
replaying from checkpointstep 2 of 4
restored full state — no work repeated
✓
send_welcome_email89ms
resumed — card not re-charged
✓
notify_slack64ms
#orders
✓
workflow complete
0 duplicate side effects
AI workflow generator
Describe it in English. Ship it in minutes.
Skip the drag-and-drop entirely. Tell OrcFlows what you need and it generates the complete workflow — nodes, expressions, data mappings, and connections.
Powered by Claude, GPT, NVIDIA NIM, or any OpenAI-compatible model
Generates fully wired nodes with correct data mappings
Everything stays editable on the canvas afterwards
Token usage and cost tracked per step, per run
"When a customer signs up, analyze their profile, route enterprise accounts to sales, and send a welcome email."
⏎ generate
Generated — 5 nodes, wired
trigger webhook · POST /signup ✓
llm claude-sonnet · analyze profile ✓
condition intent.is_enterprise ✓
slack notify #sales ✓
email send welcome · sendgrid ✓
Platform
Built for the workflows that can't afford to be wrong
Most automation tools were built for simple SaaS webhooks. OrcFlows was built for the runs that fail at 3am and matter at 9am.
Temporal-powered durability
Workflows survive server crashes, deploys, and network failures. Durable execution guarantees every step runs exactly once — no half-executed automations, no manual recovery, ever.
exactly-once
Go performance backend
A compiled Go binary, not a Node.js event loop. 65 MB idle, sub-100ms cold start, 10,000+ concurrent goroutines on a single worker. Your infrastructure bill notices the difference.
go 1.22 · compiled
Human-in-the-loop
Pause any workflow at an approval gate. Decisions route via Slack or email; execution resumes the moment someone responds.
Persistent agent memory
Agents recall prior runs, user preferences, and step history from PostgreSQL-backed memory — smarter with every execution.
Sandboxed code execution
Run Python, Node.js, or shell inside isolated Docker containers with network and filesystem controls. Safe for multi-tenant production.
Benchmarks
216,474 requests. Zero failures.
Not a synthetic hello-world. 200 concurrent Locust users for 8 minutes, where every trigger ran a live 5-step Temporal workflow — set → HTTP → transform → condition → set — with each step persisted to PostgreSQL. On a laptop.
Full API surface tested at once: auth, CRUD, triggers, executions, secrets, knowledge base
n8n's own published benchmark uses 5 virtual users on a dedicated 4 GB cloud VM — and admits 100s+ response times under load
Reproducible: the locustfile ships in the repo
Endpoint
p50
p99
req/s
Fail
GET /health
1ms
32ms
150.1
0%
GET /workflows
2ms
66ms
32.5
0%
POST /trigger5-step workflow
9ms
180ms
10.8
0%
POST /workflows
15ms
360ms
10.8
0%
POST /auth/login
63ms
200ms
10.6
0%
All 20 endpoint types
3ms
160ms
451
0%
Locust 2.31.3 · OrcFlows API + worker + Temporal OSS + PostgreSQL 16 + Redis 7, all in Docker. n8n comparison data from their published documentation.
Benchmark rigor200 vs 5
concurrent users — 40× more than n8n's own published test
Failure rate0%
across 216K requests; n8n reports 100s+ responses under load
Trigger p99180ms
under 200 users, full durable pipeline per trigger
Peak memory19 MB
Go heap under full load; Node.js stacks idle at 250 MB+
Comparison
What the alternatives can't do
We didn't build another n8n. We built what Node.js-based tools fundamentally cannot be.
● OrcFlows
n8n
Flowise
ActivePieces
Zapier
Execution engine
Crash-proof durable execution
✓
—
—
—
—
Exactly-once step semantics
✓
—
—
—
—
Human-in-the-loop approvals
✓
Limited
—
—
—
Full event history & replay
✓
—
—
—
—
AI capabilities
Natural-language workflow generation
✓
—
—
—
—
LLM token & cost tracking
✓
—
Partial
—
—
Persistent agent memory
✓
—
Partial
—
—
Deployment
Idle memory, single worker
~65 MB
~280 MB
~240 MB
~260 MB
Cloud only
Self-hosted
✓
✓
✓
✓
—
Per-task pricing
Never
None
None
None
Yes — expensive
120+ integrations
Connect your entire stack
Real, typed, authenticated connectors — not webhooks in disguise.
SlackGitHubStripeAirtableShopifySalesforceHubSpotJiraNotionLinearGmailGoogle SheetsGoogle DriveGoogle CalendarTwilioSendGridPayPalAWS S3PostgreSQLMySQLMongoDBRedisKafkaMQTTPagerDutyDatadogZendeskDiscordTelegram+ 90 more
Pricing
Simple pricing. Never per-task.
Every plan includes unlimited executions. Charging per run punishes you for automating more — so we don't.
Free
For individuals trying out durable automation. No credit card required.
Correct. Every plan — including Free — has unlimited workflow executions. Per-task pricing means your bill grows every time your automation does its job. We charge a flat monthly price per workspace, so automating more never costs more.
Can I self-host OrcFlows?
Yes. The Community Edition runs on your own infrastructure with a single Docker Compose command — API, worker, Temporal, PostgreSQL, and Redis included. Free forever, no workflow or execution limits. Enterprise adds managed self-hosted deployments with SSO and SLAs.
What happens when I hit the 10-workflow limit on Free?
Nothing breaks. Your existing workflows keep running with full durability — you just can't activate an eleventh until you upgrade or deactivate one. We never pause or delete your work.
Which AI models can I use?
Claude (Anthropic), OpenAI GPT models, NVIDIA NIM, and any OpenAI-compatible endpoint. You bring your own API keys, stored with AES-256-GCM encryption. LLM and agent nodes track token usage and cost per step automatically.
What makes execution "crash-proof"?
Every workflow runs on Temporal, which persists state at each step. If a server dies mid-run, the workflow resumes from the last completed step on restart — no duplicate emails, charges, or API calls, and no manual recovery. Exactly-once execution, not retry-and-hope.
Can I cancel or change plans anytime?
Yes. Upgrades take effect immediately; downgrades and cancellations apply at the end of the billing period. Your workflow definitions are exportable at any time.
Ship your first durable workflow today
Free plan, no credit card. Build it on the canvas or describe it in English — either way, it won't break at 3am.
