Owning an AI capability means more than hosting a model or keeping source code.
Your team needs to know how work is routed, which boundary applies, what data moved, and whether another operator can resume the result. Without that record, an “owned” system can still depend on one tool, one person, or one opaque run.
August focused our open stack on that operating record.
Work should survive the tool
AIWG continued to separate project capability from any one assistant. A project can keep its agents, skills, rules, context, and evidence in files that supported tools already understand. Discovery routes a request to the capability the project owns. Graph and mission controls make longer work visible enough to pause, resume, retry, and review.
This is how we test the portability claim ourselves. We use the same project structure across different agent tools. The native controls differ. The project’s operating intent and retained proof remain.
The August release work also strengthened setup and release evidence. Signed artifacts, checksums, dependency records, and post-publish checks reduce the gap between “the job ran” and “the release is the thing we intended to ship.”
AIWG is open source and independent. A team can adopt it without buying a service from Integro Labs.
Compute should expose the real boundary
agentic-sandbox worked on the point where an agent receives a computer and starts real work.
Fleet workloads gained clearer state and activity evidence. Recovery work makes interrupted jobs easier to reason about. Docker control moved toward safer credential-free paths. VM provisioning fixes and Celld qualification improved the stronger isolation options.
For a buyer, this is not a debate about container names. It is an operating choice. Some work can run with a lighter boundary. Some work needs a separate kernel. A useful platform should expose that choice, prove what it selected, and report enough state for an operator to act.
We do not send every task to the most expensive boundary. We match the hardware and isolation to the risk. Cost discipline is part of the architecture, not a cleanup exercise after launch.
Memory should stay portable and governed
Fortemi is the release name for the matric-memory server line. Its August work tightened the hosted path without abandoning local ownership.
Tenant controls, request admission, durable audit receipts, credential storage, outbound policy, quota checks, and job recovery make the server more legible as an operating system for memory. AWS KMS startup checks help a deployment fail closed when key handling is incomplete.
fortemi-react carried the same ownership model into the browser. Its August release added source identity for imports, approved-field search filters, final-delete receipts, and clearer Knowledge Shard profile limits. Teams can keep a local data layer while retaining proof about origin, scope, and deletion.
That distinction matters. Portable data is not just an export button. The archive needs to say what it contains, what it leaves out, and which reader can verify it.
Publishing should produce reusable truth
Pagenary moved further from a page renderer toward a publishing system.
A small site can now publish pages, route records, search data, text, JSON, and opt-in Fortemi data from one source. Tenant-local app adapters support richer views without replacing the static publishing path. The style catalog gives teams a public place to inspect and test visual packages before adopting them.
This lets documentation serve people and tools without maintaining two unrelated publishing systems. The human page remains fast and hostable almost anywhere. The structured outputs give search, agents, and downstream applications stable material to reuse.
The stack proves the delivery model
We build the tooling we deliver with, and we run it on ourselves before we run it for a client.
The August pattern is consistent:
- AIWG keeps capability and workflow evidence with the project.
- agentic-sandbox makes the compute boundary and workload state visible.
- Fortemi governs hosted memory and retains audit evidence.
- fortemi-react keeps browser memory portable and scoped.
- Pagenary publishes human pages and reusable data from the same source.
This is owned capability in practice. It can move between tools. It can run at the right cost. It can leave a record another person can inspect. It does not require every workload to become a cloud service or every decision to consume the largest token budget.
If you are deciding where owned, cost-engineered automation fits—and where it does not—book a discovery call or send a no-fee inquiry.
Read the full reports:
- AIWG August report
- agentic-sandbox August report
- Fortemi August report
- fortemi-react August report
- Pagenary August report
The open tools are available independently. Integro Labs provides the operating and licensing vehicle for teams that want help applying them; Matric remains the licensed platform tier.
Make owned capability practical.
Bring us the operating problem, brittle workflow, or automation boundary you need to make inspectable.