Design
Describe the outcome. AI Studio compiles an intent brief into a workflow you can read, with every tool, credential and approval named before anything runs.
Self-hosted · one executable · your PostgreSQL
The AI operator for your business workflows.
Janusly designs workflows with you, runs them on a durable queue, and when a step breaks it clusters the failures, replays them and shows you the evidence. One Go binary, on your infrastructure.
Docker image or a single binary. PostgreSQL 18. No control plane to rent.
- One Go executable
- PostgreSQL as the durable queue
- MCP client and server
- OpenTelemetry built in
- Self-hosted, credentials never leave
Why Janusly
Four jobs most tools split across four products.
Design, run, recover and operate live in the same runtime, so a failure on Tuesday is a workflow change on Wednesday, with the evidence attached.
Run
Every step is a row in your PostgreSQL. Idempotent starts, bounded workers, LISTEN/NOTIFY wake-ups and polling as the fallback, so nothing depends on a broker you have to babysit.
Recover
Failures are clustered by cause. Replay one, run a campaign over the cluster, or let auto-healing propose a fix that is validated on one sample first, with external writes suppressed. Recovery contracts carry fixtures that must pass before a rollout.
Operate
AI spend budgets per workflow, alert policies, a circuit breaker that pauses every entry point until you resume, public status pages, and metrics and traces out of the box.
Product
What an operator sees on a bad morning.
Three surfaces, one runtime, no diagrams: every screen below is the product over a seeded organization.
Compare
Where it sits next to what you may already run.
Integration platforms are fast to start; workflow engines are durable; Janusly is the operator on top: durable, self-hosted, and built around recovery.
| Capability | Janusly | Hosted iPaaS | Low-code automation | Workflow engine |
|---|---|---|---|---|
| Self-hosted single binary + PostgreSQL | Cluster + services | |||
| Recovery with evidence (clusters, replay, validated auto-healing) | Manual replay | Manual replay | Retries, no operator | |
| AI operator: intent brief to a readable workflow | Assistants, hosted | Nodes, not an operator | ||
| MCP client and MCP server | Client only | Client only | ||
| AI spend budgets, RED metrics and traces included | Task quotas | Add-ons | Metrics, no budgets |
Categories, not vendors.
Pricing
Run it yourself for free. Pay when you want us on call.
Self-hostedFreeThe binary, the image, unlimited workflows.
- Full runtime and Recovery Center
- Bring your own AI provider key
- Community support
TeamTalk to usFor teams that run workflows others depend on.
- Everything in Self-hosted
- WorkOS SSO and SCIM provisioning
- Managed updates and priority support
EnterpriseCustomRecovery SLAs, dedicated support, procurement.
- Everything in Team
- Audit log retention
- Named engineer, onboarding
FAQ
Questions operators ask first.
- Do I need Kubernetes?
- No. One executable serves the API, the web app, the workers and the maintenance loops, and PostgreSQL 18 is the only dependency. A container image build is part of every release.
- Where does my data live?
- In your PostgreSQL. Credentials are envelope-encrypted at rest and never appear in lists, logs or error payloads; outbound calls pass SSRF checks, DNS pinning and byte limits.
- What happens when the AI provider is down or over budget?
- The run keeps going. Every AI step has a deterministic fallback envelope, and a budget block stops the paid call, not the workflow. Anthropic is the provider today; you bring the key.
- Can it use my existing tools and agents?
- Yes. Janusly is an MCP client for your tools and an MCP server that exposes your workflows to the agents you already run.
- Is it available in Spanish?
- The product ships in English and Spanish, and so does this site.
Run your first workflow in ten minutes.
Pull the image, point it at PostgreSQL, open the Recovery Center. Nothing to sign up for.