Why we built Calyflow open source

Calyflow is an open-source recruiting OS that runs AI workflows on your own models, data, and tools. The “open-source” part is not a distribution strategy, it’s the product decision everything else follows from. The code is AGPL-3.0, the repo is public, and you can self-host the entire platform.

Here’s why we think recruiting AI, specifically, has to be built this way.

Recruiting AI is too sensitive for black boxes

A recruiting platform touches the most consequential data most companies handle: people’s careers, salaries, health-adjacent disclosures, visa situations, and on the other side, your client list and your search strategy. Now add AI to that mix, and the questions get sharp:

• What exactly happens to a CV when the AI screens it?
• Which prompts are being run against my client’s confidential brief?
• Is my candidate data training someone else’s model?

With a closed-source AI tool, the only available answer is “trust us.” With open source, the answer is “read the code”: even if you personally never will, your security team can, your enterprise client’s auditors can, and the thousand other people looking at the repo effectively do it for you. Inspectability isn’t a feature for engineers; it’s what lets a recruiting agency answer a client’s due-diligence questionnaire with documents instead of promises.

The three freedoms that follow

Being open source is what makes Calyflow’s three core principles enforceable rather than aspirational:

Bring Your Own AI. Calyflow is model-agnostic (Claude, ChatGPT, Gemini, OpenAI or Anthropic APIs, local models) on your own API key. Because the code is open, this isn’t a pricing tier; it’s verifiable architecture. Nobody can quietly insert a markup between you and your model, and when a better model ships next quarter, you switch by changing a setting, not by migrating platforms.

Bring Your Own Data. Your candidate data stays in your systems: your ATS, your CRM, your sheets. Calyflow connects to them to run workflows but does not warehouse your data. In closed platforms, your data is the lock-in; an open-source tool structurally can’t hold it hostage, because the worst case is always: take the code, self-host it, keep working.

Bring Your Own Tools. Workflows can use the sourcing and enrichment tools you already pay for. Open code means integrations are auditable, and the integration status on our homepage is driven by a single public file in the repo: what’s marked “available” is what exists, verifiably.

Why AGPL-3.0, specifically

We chose AGPL over a permissive license for one reason: the network clause. Plain MIT/Apache licensing would let anyone take Calyflow, wrap it in a proprietary SaaS, and give nothing back. AGPL says: use it, self-host it, modify it freely — but if you offer it as a service, your modifications stay open too. For a tool whose whole value proposition is inspectability, that guarantee has to survive commercialization. The improvements the community can trust are the ones that can’t be taken private.

The honest trade-offs

Open source isn’t free magic, and it’s worth saying what it costs us:

• We can’t fake progress. Roadmap slippage is public. Our homepage marks every integration “available” or “coming soon” because shipping logos before software is the kind of trust-burn an open project can’t afford.
• We compete on execution, not secrecy. Anyone can read how our workflows prompt the models. We think the moat in recruiting AI is the quality of the workflow design and the community refining it: the evidence-quoting screening method is better because it’s public and criticizable.
• Self-hosting is real work (that we’re happy to do for enterprise teams as private deployments on their own GCP). But the option existing changes every negotiation: a vendor who knows you can leave treats you differently.

What this means for a working recruiter

If you’re a recruiter rather than an engineer, the open-source label cashes out as three practical guarantees: you can answer client data-security questions concretely; your AI costs are your actual token costs with every run’s cost visible; and the tool can never charge you rent on your own data. As your team climbs the AI Adoption Ladder and more of your operation runs through workflows, those guarantees matter more each month: the more critical the tooling, the more it should be tooling you can inspect, move, and own.

The takeaway

Recruiting AI asks for deep trust: in what happens to candidates’ data, to clients’ briefs, and to your own operation. We don’t think that trust should be a promise. Calyflow is open source so it can be a property — readable in the code, enforceable in the license, and exercisable the day you want to walk away. That’s also, we’d argue, the only durable foundation for the whole BYO architecture.

The code is on GitHub: star it, read it, break it, send a PR. Or just create a free account and run your first workflow: free to start, your own API key, no credit card.