CONCEPT Cited by 1 source

Stateless agent compute

The principle of treating LLMs as stateless compute engines where all persistent state lives outside the model's context window in an external store (database, filesystem, queue).

Definition

Rather than relying on the model's context window for memory across tasks, the orchestration layer: - Persists all findings, intermediate results, and task state in external storage. - Each agent invocation receives only the minimal context needed for its hyper-focused task. - Recovery from crashes, model failures, or context exhaustion is trivial — replay the task from external state.

Cloudflare "broke the context-exhaustion bottleneck by externalizing the state entirely, treating the LLM as a stateless compute engine" (Source: sources/2026-06-18-cloudflare-build-your-own-vulnerability-harness).

Seen in

• systems/cloudflare-vulnerability-discovery-harness — SQLite DB keyed by (run_id, repo, stage) holds all state; agents are stateless workers
• systems/cloudflare-vulnerability-validation-system — same principle applied to triage pipeline

Related

• concepts/context-exhaustion — the failure mode this principle prevents
• concepts/model-agnostic-orchestration — complementary principle (model identity is also not state)
• patterns/sqlite-keyed-stage-persistence — the pattern implementation