Bolt and Aeira from Sparcle

Sparcle is the private AI plane for regulated work. It pairs Bolt, a local-first AI workspace that runs on the user's own machine, with Aeira, a governed retrieval and policy layer, to give enterprises a ChatGPT- or Copilot-class experience without sending their data to someone else's model. Everything runs inside your trust boundary: on-premises, in your VPC, or fully air-gapped. You bring your own LLM, whether that's a hosted frontier model behind your own key, a model in your VPC, or a local open-weight model Sparcle auto-detects and adopts. The result is the productivity teams expect from modern AI, under governance that HIPAA, GDPR, and DPDPA actually require.

What makes it different Governed by construction, not by policy memo. Sparcle treats governance as a chokepoint the data must pass through, not a checkbox. PII is masked at the LLM boundary so the model sees redacted context while your tools still operate on real values. Consent is enforced on the internal AI leg (DPDPA-aligned), and access is mediated by a capability and policy layer so retrieval respects each user's permissions before a single token is generated.

Retrieval that respects the ACL first. Aeira runs a hybrid retrieval ladder that filters by access control at the query level, then ranks, so users only ever ground answers in data they are already entitled to see. This is the hard part of enterprise RAG done correctly: no leakage through the retrieval side door.

Local-first, so the AI meets your data where it lives. Bolt runs as a real desktop workspace. Notes, clips, files, and search stay device-local and encrypted at rest. Drop in a file, a log, a CSV, or two documents to diff, and Bolt extracts, understands, and acts on it locally, then reaches for an LLM only when it needs to, through the masking boundary.

Bring your own model, including on-device. No lock-in to a single provider. Point Sparcle at a frontier API, a model in your cloud, or a local OpenAI-compatible endpoint it discovers automatically. On-device transcription and text-to-speech mean even voice never has to leave the machine.

Built for regulated verticals. Additive compliance packs, advisory guards for regulated domains, and provenance tracking are part of the platform, not bolted on. The same architecture serves healthcare, finance, and public-sector teams who cannot use public AI tools at all today.

Deploys the way your infrastructure demands. Three paths from evaluation to production: a self-contained trial sidecar for a single user, a single-host install, and a Helm-based cluster deployment for scale. Same governance guarantees at every size, including air-gapped.