OutcomeOps Is Now Open Source
As of today, the entire OutcomeOps platform is open source under the Apache License 2.0. Not a community edition with the useful parts held back: the same code that runs PR review, the organizational knowledge base, code maps, the MCP server, multi-region failover, and ticket-to-PR code generation. There is no license key, no paid tier, and nothing that calls home. You clone it, apply the Terraform into your own AWS account, and it’s yours.
The code is at github.com/bcarpio/outcome-ops-ai-assist, and the documentation is at docs.outcomeops.ai. This post covers why we released it, why now, and what we think it means for large organizations, particularly regulated ones, that are still working out where AI belongs in how they build software.
The intelligence layer underneath every AI tool you ship. See all the demos.
The Market Is Early, and That’s Fine
Outside of the hyperscalers and a short list of AI-native companies, most enterprises are still early in figuring out how AI engineering fits into their ecosystem. That isn’t a criticism. It’s where the market is.
The pattern in most large organizations looks similar. There’s a coding assistant rolled out to some or all developers. There’s an internal chat assistant. There’s an AI governance committee writing policy. What usually isn’t there yet is the connection between AI and the organization’s own engineering knowledge: the architecture decisions, the standards, the reasons a system is shaped the way it is. So the assistant produces plausible code that fails review for reasons it was never told about, and the review burden moves to the most senior engineers instead of going away.
That gap is what OutcomeOps was built to close. But closing it requires an organization to treat its standards as an input to AI, not just a wiki page, and most organizations aren’t there yet. They’re still deciding what the category is. Selling a platform into a market that’s still defining the problem is slow, and it puts a sales process between engineers and the thing they actually want to evaluate: whether the pattern works on their code.
Open source removes that step. A platform team can read every line, deploy it into a sandbox account, point it at twenty repositories and a folder of ADRs, and decide for themselves. Nobody has to sign anything first.
Why It Fits Open Source Anyway
OutcomeOps was always designed to run entirely inside the customer’s AWS account. Source code, ADRs, embeddings, chat history, and audit logs never left it. Inference runs on Amazon Bedrock in your account, under your IAM policies, encrypted with your KMS keys. The only component that ever talked to us was licensing.
For a security reviewer at a bank or a health insurer, that one outbound dependency was the hardest part of the review, because it was the only part they couldn’t inspect. Removing it makes the architecture simpler to approve: everything that runs is in the repository, and everything in the repository runs in your account. The Apache 2.0 license also answers the procurement questions an enterprise asks of any dependency, with a permissive license, an explicit patent grant, and no copyleft obligations.
What Large and Regulated Organizations Should Push Toward
Whether or not you ever deploy OutcomeOps, the design represents where we think large organizations, and especially regulated ones, should be heading with AI engineering:
- AI grounded in your own decisions. Your ADRs, standards, and code structure should be retrievable by every AI tool your engineers use, so generated code and reviews reflect your architecture instead of generic defaults. See why ADRs matter for AI.
- Review against those decisions, on every change. Every pull request checked against the standards that apply to it, with a verdict and a citation, whether a person or a model wrote the code.
- Your account, your boundary. No source code leaving your environment, no new SaaS sub-processor, and model choice that stays yours as models change.
- Auditable by default. Every query, retrieval, and generation logged with retention you control, exportable to your SIEM, with PII redaction before anything reaches a model.
- Built to stay up when you need it most. The platform that holds the map of your systems has to survive a regional outage. See how multi-region works.
- Knowledge that compounds. When AI gets something wrong, the fix should be a better standard that every future task inherits, not a one-off correction in one chat window. See from fixing code to teaching systems.
Those are the requirements. OutcomeOps is one working implementation of them, and now anyone can read it.
What’s in the Repository
- PR review: six grounded checks on every pull request (ADR compliance, breaking changes, architectural duplication, test coverage, README freshness, license compliance), a PR summary, and cross-repo structural impact, on GitHub, GitLab, and Azure DevOps.
- Organizational knowledge base and chat: retrieval over code, ADRs, and the systems your organization already writes in, including Confluence, Jira, SharePoint, OneDrive, Outlook, Teams, and database schemas.
- Code understanding: code maps and a cross-repo symbol graph for callers, importers, and implementers.
- An MCP server that gives Claude Code, Cursor, Kiro, and other MCP clients the same organizational context.
- Enterprise controls: workspaces, roles, Entra ID sign-in, audit logging, SIEM export, cost tracking and budgets, PII redaction, and KMS encryption.
- Deployment: Terraform for a private, single-region install, with optional active/passive multi-region.
- Optional code generation: assign a ticket and get a pull request with tests, which then goes through the same review.
The full breakdown is on the features page.
The repository is also a worked example of the approach. It carries its own architecture decision records, and every pull request to it is reviewed by the platform itself, against those ADRs. If you want to see what a codebase written for AI-grounded review looks like, read the repo.
License Compliance
docs/lambda-list-recent-docs.md
New file docs/lambda-list-recent-docs.md has no copyright header. ADR-008 says new files missing copyright get a WARN, not a FAIL. No prohibited or review-required licenses were found. The file is Markdown documentation rather than first-party source code, and ADR-008 notes that headers are being added repo-wide by a later mechanical sweep, so this is low severity.
What to Expect From Here
Being direct about this matters for anyone deciding whether to build on it:
- It’s maintained. Bugs get fixed, issues get answered, and pull requests are welcome. See the contributing guide in the repository; contributions are signed off under the Developer Certificate of Origin.
- Support is community support. Questions and bug reports go through GitHub Issues. There is no hosted version, no paid support plan, and no implementation service.
- The name is a trademark. The Apache license covers the code, not the OutcomeOps name or logo, so a fork you redistribute needs its own name.
Getting Started
You need an AWS account, Terraform, Docker, and the AWS CLI. A first deployment takes roughly 30 to 45 minutes, most of it building container images and Lambda layers. The documentation walks through it, including how to start if your organization has no ADRs yet.
Start small: one sandbox account, a handful of representative repositories, and whatever standards you already have written down. Turn on PR review first. It’s the fastest way to see whether your standards are clear enough for AI to apply, and if they aren’t, that’s worth knowing on its own.
OutcomeOps: The Future of AI Engineering
Opens Substack in a new tab to confirm. No spam — unsubscribe anytime.
Read It. Deploy It. Decide for Yourself.
OutcomeOps is open source under Apache 2.0 and runs entirely in your AWS account.
Related reading
- Features — every capability, by area.
- AI Coding Tools for Regulated Industries — why deployment model dominates the compliance review.
- AI-DLC Fixed the Process. The Knowledge Is Still Local. — where OutcomeOps fits next to AWS’s own AI development lifecycle.