Back to home
    Buyer guide
    Fit

    Is ReDevOps right for us?

    redevops.io

    A straight answer to a buying question: what ReDevOps is for, what it costs you to run, and how to tell if you're ready for it.

    What problems justify deploying it

    ReDevOps earns its place when AI stops being a single feature and starts being infrastructure. You probably have a real case for it if two or more of these are true:

    • Several AI workflows, not one. You're running (or planning) multiple agents/assistants — support triage, a sales copilot, an internal ops bot, a data Q&A tool — and each one reinvents retrieval, planning, and guardrails badly.
    • Multiple data sources behind those workflows. Answers have to draw on a mix: your SQL databases, a document store, a knowledge/graph layer, third-party APIs, internal tools. No single RAG pipeline covers all of them well.
    • Consequential actions. The AI doesn't just chat — it issues refunds, changes billing, files tickets, moves money, touches production. Those actions need approval before they commit, an audit trail after, and the ability to reproduce exactly what happened and why.

    If you have all three, the pain is usually the same: answers that change between runs with no explanation, no one who can prove what an agent did last Tuesday, and cost that grows faster than the value.

    What it integrates with

    ReDevOps is a runtime layer, not a rip-and-replace platform.

    • It wraps what you already have. Existing agents and frameworks become capabilities the runtime can call, govern, and record — you don't rewrite them.
    • It sits between your apps and your models/data. Your applications talk to ReDevOps; ReDevOps brokers the models, RAG, SQL, graphs, APIs, and tools underneath, applying planning, budgeting, and approval gates in the middle.
    • The reference apps are built on open-source cores. The example deployments wrap real OSS systems — Lago (billing/metering), ERPNext (ERP), Chatwoot (support), and others — so you're integrating with familiar, self-hostable software rather than a closed stack.
    • Model- and provider-independent. Bring your own models — hosted API providers or self-hosted open weights. Nothing is pinned to one vendor.

    What the first deployment includes

    A first install is deliberately small and real, not a proof-of-concept slide:

    • Helm/Kubernetes install of the reference system — the Python runtimes (Discovery, Mission, Context, retrieval) deployed to your cluster.
    • One connected workflow/capability wired end to end against a real data source, so you see the full path work once.
    • The operating console — the screen where a person launches, watches, and approves missions and agents.
    • Approval gates + a replayable ledger — consequential actions pause for human sign-off, and every run is written down in a tamper-evident record you can replay.

    Open source vs enterprise

    • Open source, today. The Python runtimes are AGPL and self-hostable right now. You can deploy the reference system, run real workflows, and keep everything in your own cluster with no license conversation.
    • Enterprise, on request. The Go (zero-dependency) and Kotlin/JVM runtimes are the enterprise layer — the same contracts, implemented for high-throughput and JVM/Spark environments — available on request.
    • What enterprise adds around the runtime: identity & access (principals, ABAC, tenant isolation), metering, and fleet/governance operations for running many missions across teams.

    You can start entirely on the open-source Python stack and only reach for the enterprise layer when scale, JVM/Go environments, or multi-team governance demand it.

    Expected effort

    Be honest with yourself about ownership and infrastructure before committing:

    • Who owns it. A platform- or DevOps-minded team — people comfortable with Kubernetes, Postgres, and operating a service. It is not a no-code tool.
    • Infrastructure. A Kubernetes cluster and Postgres with pgvector. Optionally a GPU if you want to self-host models rather than call a hosted provider.
    • Adoption effort. Incremental — no full rewrite. Wrap one existing workflow first, prove the loop (plan → approve → run → replay), then move more of your planning and execution into the runtime over time. You adopt it one capability at a time.

    Probably a fit / probably not yet

    Probably a fit if you have:

    • Several AI products or workflows sharing infrastructure
    • AI that takes consequential actions (money, production, customer records)
    • Audit / compliance needs — you must prove what happened and reproduce it
    • Cost or consistency pain — spend growing faster than value, answers that drift between runs

    Probably not yet if you have:

    • A single chatbot or one prompt-in, text-out workflow
    • No consequential actions and no audit requirement — a lighter tool will serve you better today

    Ready to look at the mechanics? See the deploy page for how a first install goes, or read the plain-English overview for what each part actually does.