Dated, source-linked observations imported from the QWG AI archive.
These are evidence records, not finished editorial conclusions.
Across 1332 observations, most signals relate to model behavior, evaluation discipline, and real-world agent operations. Strong evidence clusters around tooling, memory, routing, and productized agent execution.
Teams are moving from experiments to systems. The emphasis is on verifiable behavior, operational safety, and measurable outcomes—especially in agent memory, evaluation rigor, and forward-deployed engineering.
Earendil announced Pi 1.0 as a stable release of its minimal, extensible agent harness, with Codemode supporting MCP and non-LLM models, virtual-model extensions, deferred tool loading, cache warming for Anthropic models, and mid-conversation system-message changes.
Cursor's model documentation now lists GLM 5.3 Flash as a selectable model under Cursor Settings > Models.
Modal introduces Sidecars for Sandboxes Modal announced Sidecars a beta feature for its Sandboxes product on October 1 2026. Sidecars are isolated containers that run alongside a main Sandbox on the same host intended to separate trusted services such as
Modal makes Clusters generally available On October 1 2026 Modal announced that Modal Clusters are generally available to all workspaces. The company describes Clusters as groups of containers for multi-node workloads launched through the modal.clustered decorator. Its example requests four
LlamaIndex announced Extract v2.5, an update to its schema-based document extraction agents.
Vercel says Microsoft AI’s MAI audio models are now available through AI Gateway.
Cloudflare has introduced Workers KV Instant, powered by Quicksilver.
Cloudflare has announced K2, a serverless event-streaming service built directly on R2 object storage.
Amazon Science published “Graph-centric agentic intelligence,” describing an approach that combines network digital-twin graphs, cascaded graph analytics, and agentic orchestration for root-cause analysis.
Ai2 has released Olmo-core 3, an open training framework redesigned for large mixture-of-experts (MoE) models.
No raw signals match these filters.
Primary indexes and programmatic access to this dataset.
Loading the privacy-safe route aggregate…
Open the JSON contract