The governed memory layer for AI agents
Critical Context. Controlled Recall.
Useful memory reaches the model. Restricted context does not. Gluevenir gives each authorized persona the context their work needs, then attaches independently verifiable proof.
Agentic runtime controls
Five ways to keep memory moving safely.
An AARM-inspired gateway decides before supported memory actions execute. Selected AIUC-1 implementation evidence covers scoped access, tenant isolation, input and output checks, bounded tools, and content-safe audit events. Explore the complete mappings.
Interactive decision path
Ask once. Watch memory earn access.
Choose a role, start from a realistic training scenario, and watch the agent assemble the right context for the job.
Signed Recall ReceiptContent-safe proofExample only · send the request to verify
- Receipt ID
- Decision / reason
- Included memory IDs
- Content hashes
- Action hash
- Policy hash
- Exclusions
- Signing key
- Signature
Synthetic Memory Library
Browse what the demo can remember.
HelixCure Synthetic Labs is advancing HX-17, a wholly fictional investigational monoclonal antibody intended to treat a rare complement-mediated inflammatory disorder. Its simulated Phase 1 program follows protocol HC-HX17-101 v1.3, formulation HX17-F3, enrollment at Harborlight Research Center Site SYN-03, and partner communications with Argent Bridge Biologics. Browse its synthetic protocol records, stability work, cohort updates, approvals, partner releases, and lifecycle events. Persona focus organizes the library only; the Memory Action Gateway decides runtime access independently.
Persona focus: an editorial browsing cue only. It never grants memory access or changes a gateway decision.
Loading the wholly fictional memory catalog…
Public realism references
Authoritative public sources inform generic terminology and record structure only. HelixCure, HX-17, every person, protocol, result, decision, and approval are fictional; no source endorses this demo.
Live, content-safe evidence
See governed memory at work.
Explore aggregate outcomes, fixed persona views, and request traces. Telemetry contains bounded stages, decisions, counts, verification flags, and duration. It never contains prompts, answers, memory content, or customer identifiers.
The embedded viewer becomes available in local Compose or after the dedicated hosted observability URL is configured.
AARM-inspired, deliberately bounded
The gateway maps AARM-inspired evidence to supported memory-mediated actions only. This demonstration does not claim AARM conformance, certification, or universal agent interception.
AIUC-1 evidence, not certification
The prototype produces evidence relevant to selected AIUC-1 controls. It is not AIUC-1 certified, conformant, assessed, audit-ready, or a substitute for organizational and legal controls.
Development-key trust model
Receipts are signed, content-bound, and independently verifiable within the documented development-key trust model. That boundary does not protect against an administrator controlling both storage and key.