What we're building—and what we're learning.
Engineering notes, public-specification work, product releases, research, and company announcements from XSI.
XSI Reckoning: 40× lower input cost for reasoning
At turn 50 of XSI's controlled research comparison, full-transcript reasoning required 11,450 input tokens. Reckoning's bounded-memory approach required 286: a 97.5% reduction, before tool overhead.
Article2026-09What the Memory Continuum remembers
The Memory Continuum architecture connects current standing, durable records and execution state so work can resume with the evidence and authority it needs.
Company Log2026-09Illia Artiukh joins XSI as Co-Founder and Head of Engineering
XSI expands its founding leadership team as the company moves from architectural proof into a broader phase of product engineering, integration, and production deployment.
Blog2026-08Made with XSI: A Mark That Can Be Checked
A record of origin and a verifiable signature can disclose AI participation and bind that statement to an exact artifact.
Press release2026-07XSI-AIMS for the Microsoft Agent Framework is now available in Microsoft Marketplace
Microsoft customers can discover XSI-AIMS™ for the Microsoft Agent Framework through Microsoft Marketplace and contact XSI about deployment.
Press release2026-07XSI introduces XSI Apprentice, a governed programming-by-demonstration tool, in preview
XSI Apprentice is in preview. Its design connects a demonstrated task to a workflow approved by a second person, with human review of every run’s work product.
Press release2026-06XSI previews conversational voice support for LodeStone and XSI-AIMS agents
Agents can join a meeting by speech under the same XSI-AIMS authority chain, Witness Layer, and shadow governance as every other agent. Preview access is by invitation and waitlist as of June 2026.
Blog2026-06Agent governance needs a runtime layer
Agent governance is an architecture question: a runtime layer that supervises what an agent intends, plans, and does, on any model or cloud or none at all. XSI-AIMS is that specification, published in the open.
Blog2026-06Why Agentic AI Needs a Governance Specification
An agent framework runs the workflow. A governance contract defines authority, policy and the evidence that should accompany its actions across deployments.
Blog2026-06Biodiversity of Virtual Lifeforms
A long-running agent can combine specialist models, memory systems and tools under a shared governance contract. Each component can evolve without making the whole system one indivisible model.
Press release2026-06XSI-AIMS v2.0 specification published for public review
Extended Systems Intelligence published XSI-AIMS v2.0 for public review in June 2026. The open specification defines a governance contract for agentic systems across models, runtimes and deployment environments.
Blog2026-06XSI Compass: Route Model Calls by What the Work Requires
Cost is one input to model selection. Required capabilities, deployment policy, context size and latency determine which routes are eligible.
Blog2026-06Why XSI-AIMS Defines Different Kinds of Memory
Current state, historical events, procedures and future commitments serve different jobs. XSI Reckoning carries the active task, while XSI Atlas retains durable records.
Milestone2026-05AppSource listing approval for XSI-AIMS for the Microsoft Agent Framework
Microsoft accepted the XSI-AIMS for the Microsoft Agent Framework listing for AppSource in May 2026.
Press release2026-06XSI-AIMS™ for MAF is live on AppSource.
A substrate runtime of XSI's open governance specification, now available as a Contact-Me listing on Microsoft AppSource for teams building on the Microsoft Agent Framework.