Purvakalpa Studio

Docs / Storyline

Roadmap

What is in progress, designed or planned but not shipped, kept apart so the other pages only describe what is in Storyline 1.2.1 today.

Everything on the other pages is in Storyline 1.2.1, the version on the App Store. This page lists what is not, so the line between built and planned stays clear.

ItemStatusNotes
Family-voice narrationIn progressBuilt for release 1.3.0, which is not on the App Store yet. A scene, or a whole chapter in reading order, is read aloud in a family voice served by a companion voice keepsake app on your own Mac, over your own Wi-Fi, on an encrypted, authenticated connection with no server and no account. Pairing takes a six-digit code, once. Every clip lands as a pending, reviewable asset marked as an AI voice, with the voice and the Mac that made it on record. With no Mac in reach it refuses and says what to do. In privileged projects, attaching a clip is refused until narration audio can use the encrypted asset path
An audiobook in your own voiceIn progressTurning a draft into an audiobook narrated in your own voice, generated on your device. Described at 1.2.1 as early, but already working. Read aloud with Apple voices, led by your own Personal Voice, already ships
Your devices as a private clusterIn progressDescribed as a bet, not a feature. iPhone and iPad would be the writing surface and the Mac the workshop for heavy passes, such as a critique of a whole manuscript. A small, deterministic rule engine would pick one of four answers for each task: run it here, hand it to a stronger device over the local network, queue it in your iCloud for whichever device wakes first, or refuse and say exactly why, never a silent downgrade to a weaker model. The encrypted transport between your devices exists today; the cluster on top is a phased build
Prefix cache for the knowledge baseIn progressA KV cache for the project’s reference material, so the on-device model does not have to reprocess it every turn. The cache store, manifest and hashing are built; wiring them to the live model waits on upstream mlx-swift support
Anonymising relay for web searchDesignedOblivious HTTP (RFC 9458): a relay sees your address but not the query, and a gateway sees the query but not your address, which holds only if the two do not collude. For web search and fetch only, never model calls. Off by default, and a relay that is down never falls back to a direct send. Until it exists, the Anonymise web searches option keeps research on the device
Matter lifecycleDesignedRetention set as policy data, closing and sealing a matter, cryptographic destruction with a Secure Enclave-signed certificate and legal holds, and a local conflict check when a matter is created, with the check itself logged
Share extensionDesignedCapture from other apps into an encrypted spool, routed to a project when you review it
Pseudonymised contextPlannedAn alias map so a model never sees real names, with aliases swapped back locally on output
Leak guard tier 2PlannedA small local classifier for the grey zone between general knowledge and confidential particulars
Evidence-locked draftingPlannedA mode in which the model may assert only what a clip supports
Off-device audit anchors and export provenancePlannedNotarised audit heads anchored off the device with RFC 3161 timestamps, and signed export manifests that state how much of a document is your own writing
External MCP serversPlannedLocal tools are already described through an internal MCP-shaped seam with no external transport; connecting to external MCP servers would reuse it
Composable master agentsPlannedA master agent’s set of specialists is defined in code today. A saved, editable composition and a composer would let you build your own

Have a request? Use the feature request form.