An open standard for agentic systems.
Substrate-agnostic. Model-agnostic. Deployment-agnostic.
AIMS is an open governance specification authored by XSI to keep agentic systems supervised, accountable, and sovereign — regardless of which models, runtimes, or environments deliver them.
Four altitude layers.
AIMS organizes agentic operation across four operational altitude layers. Every conformant implementation maps each piece of agentic work to one of these layers, with conformance attestation at each boundary.
What is being asked.
The principal's request, the authority posture, and the policy envelope under which the request is admitted.
How it gets done.
The plan synthesized from intent, with shadow-governance observers attached at planning time.
Construction.
The plan is constructed into concrete actions — tool calls, model calls, sub-agent dispatch — with witness layers attached at every boundary.
Execution.
The constructed actions execute under continuous supervision. The gateway enforces policy; failures escalate; every step is recorded for attestation.
Three functional modes.
Every conformant implementation supports three modes. They run concurrently across all four altitude layers, observing each other through the Radial Gateway.
Produces. Synthesizes plans, drafts text, proposes tool calls, generates the candidate work.
Audits. Checks the candidate work against policy, conformance, safety bumpers, and the authority posture established at L-Intent.
Reconciles. Merges generative output with constraining feedback, resolves conflicts, and produces the version of record that proceeds.
The instruments.
Adopting AIMS gets you both reference tools immediately. Both are OSS, free, and tied to the AIMS agent registry.
AIMS Atlas
Memory routing harness across the five AIMS memory categories — PMS, KRS, SMS, TMS, EMS. One interface; swappable backends.
Atlas docs → v1 · liveAIMS Compass
Model routing tool. Register your provider keys; Compass picks the cheapest model meeting each call's requirements, with adversarial-challenge for safety-critical calls.
Compass docs →One registry for conformant agents.
The AIMS agent registry is the canonical record of every conformant agent, the providers it consumes, and the AIMS-conformance attestation it signs. Atlas and Compass register their decisions to the registry's audit log.
The what, not the how.
AIMS conformance means every required feature is present. How each feature is delivered remains the implementer's intellectual property. This is the same pattern as SQL → Postgres / MySQL, HTTP → nginx / Apache, POSIX → Linux / macOS / BSD.
XSI's three reference deployments — LodeStone, AIMS Advisor, XSI-AIMS for MAF — each claim AIMS conformance through their own proprietary mechanisms. Third-party implementations on any substrate, any model, any environment, are equally legitimate.
An open standard, open governance.
AIMS is authored by XSI but community-owned by intent. Submissions to candidate standards bodies — NIST, IETF, W3C, OASIS, and MCP — are planned after the June 2026 release.
XSI will continue to maintain the canonical spec, the reference toolkit, and the agent registry while the standards process unfolds.