Find meaning across private application data.
Primordial SDKAlpha
The application layer for on-device AI
Build grounded, multimodal, and agentic experiences across search, generation, chat, voice, and vision. Primordial handles the models, runtimes, retrieval, memory, and device infrastructure beneath them. Build grounded, multimodal, and agentic experiences across search, generation, chat, voice, and vision. Primordial handles the models, runtimes, retrieval, memory, and device infrastructure beneath them.
The Missing layer.
The Missing layer.
Primordial simplifies building with on-device AI.You focus on your outcome instead of the infrastructure.
-
Is the required model available?
-
How is local data prepared?
-
How do we retrieve relevant context?
-
What should enter the prompt?
-
Can this run reliably on the device?
Respond with evidence from sources your application trusts.
Turn speech into private, responsive application actions.
Build conversations with memory and relevant context.
Coordinate tools, context, and execution on the device.
Primordial SDK
- Models
- Data
- OS
- Hardware
Many challanges
The building blocks AI applications actually need.
Move beyond chat demos.The fastest path from on-device intelligence to production apps and enterprise workflows.
Run models reliably.
Manage availability, background downloads, loading, shared access, memory pressure, and recovery.
Search your data by meaning.
Build local collections and retrieve the right context by meaning, not keywords.
Answer with evidence.
Generate answers from your data and return citations to the sources used.
Turn prompts into workflows.
Summarize, extract, classify, or define a generation task for your product.
Build chat that knows your data.
Create model-only or grounded conversations with local context and citations.
Give your app a voice.
Record and transcribe audio locally with managed session and model readiness.
One SDK between intelligence and your product.
Record speech, transcribe it, and summarize the transcript into three bullet points without assembling separate speech, model-lifecycle, and generation stacks.
import Primordial
let primordial = PrimordialClient()
try await primordial.activate(.evaluationKey("pk_eval_your_key_here"))
for try await progress in primordial.ai.makeAvailable([
.voiceInput,
.generation
]) {
print(progress.fractionCompleted ?? 0)
}
let recording = primordial.recording()
try await recording.start()
let result = try await recording.stop()
let summary = try await primordial.summarize(
result.transcription.text,
style: .bulletPoints(maximum: 3)
)
print(summary)
Why on-device
Run intelligence where
the data already lives.
On-device AI keeps data private and costs predictable, so
you can ship without token billing, usage caps, or API bills.
No per-token bill
Local workflows do not add a cloud inference charge for every user interaction.
Privacy built-in
Prompts, files, audio, and generated output can remain on the user’s device.
Fast and offline
Remove the inference server round trip and keep core workflows available after required models are ready.