Describe what you want. CorvinOS builds the workflow. Export it as a portable AWPKG, share it with a partner, and it runs on their CorvinOS — with their data, under their rules. Every run is hash-chained and EU-compliant.
Pipelines in CorvinOS are not configuration files or cron jobs. They are first-class programs with names, versions, inputs, outputs, and logic — running on an operating system that provides the infrastructure.
A workflow has a name, a version, defined inputs, and verifiable outputs. It can branch on conditions, loop over data, and call external services. CorvinOS runs it on demand or on schedule, tracks every execution, and keeps a full audit trail — just like an OS manages processes.
Like an operating system runs applications, CorvinOS runs workflows. Different machines, same result. The workflow brings its own logic; CorvinOS provides agents, databases, secrets, compute, and compliance — the infrastructure every workflow needs to run.
Export any workflow as an Agentic Workflow Package (.awpkg). Send it to a partner organization. They install it on their CorvinOS. It runs with their data, their agents, and their secrets. Nothing from your side leaks. It works like software distribution — but for AI workflows.
You don't need to build pipelines step by step. Tell CorvinOS what you need — it generates the workflow, you review and confirm, and the program is ready to run.
Tell your agent what you need in plain language — no YAML, no pipeline editor, no configuration UI.
The system creates a complete pipeline: steps, triggers, conditions, data source connections, and notification targets — wired together automatically.
Review the generated workflow in the console. Edit any step. Save it. Export as AWPKG whenever you want to share it. The workflow is yours — permanently stored, versioned, and audited.
An AWPKG is a signed, portable file that bundles a complete workflow definition. Transfer it like software. Install it on any CorvinOS instance anywhere in the world.
An AWPKG bundles the entire workflow definition: steps, triggers, conditions, schema, and version metadata. Transfer it like you'd transfer a document — email, USB, SFTP, marketplace.
Every package is signed by the author's key. The installing CorvinOS verifies the signature before execution. A tampered package is rejected. You know exactly where the workflow came from and that it hasn't been modified in transit.
Install on your own server, a partner's server, or a cloud instance. The workflow connects to local data sources, local agents, and local secrets — nothing from the original author is required at runtime.
Build a Monday briefing once. Export it. Every department installs their own copy, pointing at their own database. One workflow, dozens of installations, each running independently and audited separately.
Every pipeline is made of steps. Mix and match these six types to build workflows that think, query, compute, integrate, and communicate.
The AI reasons about the input, produces text or structured JSON, and can call tools. Use for decisions, summaries, classifications, and drafts.
Query a connected PostgreSQL, MySQL, or SQLite database. Results flow as structured data to all downstream steps.
Kick off a heavy computation — statistical analysis, ML inference, grid search, or data transform — without blocking the conversation.
Call any REST API with data from previous steps. Integrate with PagerDuty, Jira, Slack, GitHub, email gateways, or any webhook.
Delegate a signed task to another organization's CorvinOS. Wait for their cryptographically verified response, then continue with the result.
Post results to Discord, Slack, Telegram, email, or any webhook endpoint. Target specific channels, threads, or users.
Workflows start on your terms — on a fixed schedule, when someone sends a message, when an external system pings you, or when another agent finishes a task.
Run on a cron expression. Every weekday morning, once a week, every hour.
0 7 * * 1-5A team member types a keyword or command in a connected channel — the workflow starts immediately.
pattern: "!report"HTTP POST from any external system — GitHub push, Jira update, Prometheus alert, payment event.
POST /hooks/pipeline-idWhen another agent in your organization produces output that matches a condition you define.
on: agent.output.severity == "high"When a partner organization's agent finishes a delegated task and sends back a signed response.
on: a2a.response_receivedStart from the console, via a chat command, or through the REST API whenever you need it.
POST /v1/pipelines/:id/runFour example workflows — all exportable as AWPKG. Install them as-is, or customize for your data sources and channels.
Trigger: Every weekday at 07:00
PostgreSQL: open tasks due today
Summarize tasks and highlight blockers
Post to Discord #morning-standup
Trigger: Webhook — Prometheus alert fires
Classify alert: service, severity, likely cause
If severity == HIGH → create PagerDuty incident
PagerDuty: create incident
Draft incident summary with recent deployment context
Post to #incidents with full context and Jira link
Trigger: Every Friday at 16:00
Delegate to Research partner: top 5 AI papers this week
Statistical analysis of citation counts + trend scores
Write a 3-paragraph executive summary
Send digest via email API
Trigger: Message posted in #support channel
Classify: type (bug / question / feature) + sentiment
Knowledge base: similar resolved tickets
Draft reply using knowledge base context
Reply in thread with draft + optional Jira ticket
EU compliance is not a checkbox you tick at the end. In CorvinOS, it is built into every pipeline step — automatically, without configuration.
Every pipeline step is recorded in a tamper-evident hash chain. A modified or missing entry breaks the chain and triggers a CRITICAL alert. You can prove exactly what ran, when, with what input, and what the output was.
Full records of all processing activities are maintained automatically. No manual logging required. The audit log is on your server, under your control, readable offline, and verifiable without a third party.
Pipeline execution history, step-level traces, failure reasons, and intervention points are always visible in the console. You can inspect any run, re-trigger with corrections, or halt a workflow at any time.
Workflows run on your infrastructure. No step sends data to any third party unless you explicitly configure an HTTP call. For maximum isolation, run AI inference locally with a local model — zero egress, zero cloud dependency.
Open-source, self-hosted, EU AI Act enforced by design. Describe a workflow, export it as AWPKG, and run it anywhere.