ai-memory v1.0.0 · Acceptance Campaign — Live

Release — · certified pg 18.6 / AGE 1.8.0 / pgvector · —
GLM-5.3-Flash swarm/hive → certified data tier, encrypted in transit (mTLS) · Fable authors · GLM executes · honest-by-construction

connecting…

—

dimensions green

North Star · the Fortune-500 / Federal bar

Live swarm · GLM-5.3-Flash NHIs under test

Agent fabric

each cell is one NHI exercising ai-memory over mTLS — color = live state

working idle / ready recovered from cold error standby

Throughput

ai-memory operations / second — live stream

Federation & coordination

Federation topology

certified data-tier modules · mTLS-encrypted links · live health

A2A message flow

agent-to-agent coordination primitives — live rate & mix

Runtime performance · ai-memory

Startup latency

MCP-ready time — the #3329 non-blocking-boot guarantee (backfill never blocks init)

Operation latency

per-op p50 / p95 / p99 over the certified data tier

Performance envelope

Per-module capacity ramp

concurrent NHIs vs sustained throughput — the knee = agents one certified module handles

Latency vs budget

p50 / p95 / p99 against PERFORMANCE.md

Feature coverage

% of the drivable tool surface exercised end-to-end

Module scaling

sustained agents as certified modules are added (federation)

Security & resilience

Memory-poisoning defense

adversarial "bad memories" injected vs infections that spread — red-team

Hard stop/start continuity

kill cycles → recovered mission·goals·objectives, resumed exactly where it left off

Test dimensions

Live event feed

Configuration under test · every agent mirrors this TOML

Fleet-standard ai-memory configuration — byte-identical on every node & agent wired into ai-memory (only db and key-file paths differ per node). Everything hits OpenRouter; Ollama is not used on any node.

—

Fable 5 (conductor) · realtime via Cloudflare · a dimension turns green only when measured · state pushed by the f2 harness.