Narrative Council

# AI and Provider Disclosure **Effective date:** September 1, 2026 **Last updated:** September 1, 2026 Narrative Council is an AI-only service operated by Hajar Labs LLC. It coordinates specialized AI systems and infrastructure to produce a Strategic Communications Playbook. There is no promised human review. AI outputs can be incomplete, inaccurate, or unsuitable for a particular context; you should validate important facts and obtain appropriate professional advice before acting. ## Current processing roles The following active provider categories may process information needed to provide the service: - **Google Gemini:** bounded image or audio source extraction when local extraction is not supported. - **AssemblyAI:** client-requested voice transcription, including browser-to-provider streaming using a short-lived session token. - **Cloudflare Workers AI and Vectorize:** retrieval embeddings, tenant-scoped vector search, and bounded fallback execution of approved DeepSeek and GLM model workloads when a preferred transport is unavailable. - **Cloudflare R2:** storage of uploaded originals and generated exports. - **Cloudflare AI Gateway:** an approved transport, timeout, or observability layer for AI requests when configured. - **OpenRouter:** the primary transport for version-pinned strategic reasoning and final-quality review. Each request is restricted to an exact allowlisted model-hosting endpoint, disables provider fallback, requires zero-data-retention routing, and denies provider data collection. - **Tavily:** public-web search, extraction, and research for bounded public queries. Private uploaded content is not intended for Tavily queries. We minimize the information sent to a provider to what is needed for the assigned task. We do not expose raw prompts, hidden reasoning, raw provider responses, or operational traces in client surfaces. Provider terms, account settings, product configuration, and data types can affect processing location, logging, retention, and model-improvement practices. Accordingly, we do not make a universal U.S.-only processing, zero-retention, or no-training claim. See the Subprocessors page and Privacy Notice for more information. OpenRouter states that prompt and completion logging is opt-in and that it retains request metadata such as token counts and latency. Narrative Council additionally sends per-request zero-data-retention and data-collection-deny controls. The exact downstream model-hosting endpoint also processes the bounded work packet and must match the immutable provider allowlist recorded for the engagement; its location depends on that endpoint. **Not current default paths for new public engagements:** direct OpenAI, DeepInfra, and the direct DeepSeek API are disabled-by-default compatibility or emergency paths unless and until an explicitly authorized current provider profile and disclosure permit their use. Approved DeepSeek and GLM model workloads may be served by Cloudflare Workers AI under the fallback controls above. Historical compatibility work may have different provider provenance.