Tanvrit Compute SDK

> The com.tanvrit:compute Kotlin Multiplatform module — consumed by the > compute portal, Tanvrit AI, Tanvrit Market, and any downstream app.

Installation

Add the Maven artifact to your build.gradle.kts:

dependencies {
    implementation("com.tanvrit:compute:0.8.4")            // common
    implementation("com.tanvrit:compute-jvm:0.8.4")        // JVM/desktop targets
    implementation("com.tanvrit:compute-android:0.8.4")    // Android
}

Then ensure your settings.gradle.kts includes the Tanvrit Maven proxy (or GitHub Packages with credentials):

dependencyResolutionManagement {
    repositories {
        maven { url = uri("https://maven.tanvrit.com") }  // Cloudflare Worker proxy
        // or: maven { url = uri("https://maven.pkg.github.com/tanvrit/compute") }
    }
}

Koin setup

import com.tanvrit.compute.di.computeModule
import org.koin.core.context.startKoin

startKoin {
    modules(computeModule)  // registers job, node, execution, template, credit, schedule
}

Canonical feature layout

The SDK mirrors the /sdk/customers layout exactly. Each sub-feature exposes:

feature/<name>/
├── model/         @Serializable DTOs and domain types
├── event/         route path constants
├── network/       I<Name>Network interface + <Name>Network Ktor impl
├── repository/    MutableStateFlow holders for reactive UI state
├── handler/       Koin bridge linking repo + viewModel
└── viewModel/     AppViewModel subclass with actions + StateFlow properties

Sub-features shipped in com.tanvrit:compute

| Sub-feature | Package | What it does | |--- |--- |--- | | job | com.tanvrit.compute.feature.job | submit, retrieve, cancel, retry, list, bulk, stats | | node | com.tanvrit.compute.feature.node | list, drain, pause, resume, delete, rotate-key | | execution | com.tanvrit.compute.feature.execution | per-attempt execution history by job / node | | template | com.tanvrit.compute.feature.template | list, create, submit-from-template | | credit | com.tanvrit.compute.feature.credit | balance + ledger (free-tier model) | | schedule | com.tanvrit.compute.feature.schedule | cron-triggered recurring jobs |

Submit a job (DSL)

The compute-engine module provides a type-safe builder:

import com.tanvrit.compute.engine.dsl.computeJob
import com.tanvrit.compute.feature.job.viewModel.ComputeJobViewModel
import org.koin.java.KoinJavaComponent.get

val vm: ComputeJobViewModel = get(ComputeJobViewModel::class.java)
val req = computeJob {
    docker("ghcr.io/ggerganov/llama.cpp:server-cuda")
    env("MODEL", "llama-3-70b")
    gpu("cuda")
    memory(24_576)
    priority(9)
    tag("inference")
}
vm.submitJob(req) { success -> println("submitted=$success") }

Watch a job's live logs

vm.repository.liveLogs.collect { lines ->
    // bind to a LogViewer composable
}
// open the SSE stream from the server

The portal's LogViewer composable (/compute/composeApp/.../component/LogViewer.kt) subscribes to vm.liveLogs and appends each new line as it arrives from the /api/compute/jobs/{jobId}/logs/stream server-sent-event endpoint.

List nodes with filters

import com.tanvrit.compute.feature.node.model.ListComputeNodesRequest
import com.tanvrit.compute.feature.node.viewModel.ComputeNodeViewModel

val nodeVm: ComputeNodeViewModel = get(ComputeNodeViewModel::class.java)
nodeVm.listNodes(ListComputeNodesRequest(businessId = "biz-123", status = "ONLINE")) { ok ->
    if (ok) println("got ${nodeVm.allNodes.value.size} nodes")
}

Schedule a recurring job

import com.tanvrit.compute.feature.schedule.model.CreateComputeScheduleRequest
import com.tanvrit.compute.feature.schedule.viewModel.ComputeScheduleViewModel

val scheduleVm: ComputeScheduleViewModel = get(ComputeScheduleViewModel::class.java)
scheduleVm.createSchedule(
    CreateComputeScheduleRequest(
        name = "Nightly backup",
        cronExpression = "0 3 * * *",   // every day 03:00 UTC
        jobType = "COMMAND",
        command = "backup.sh --target s3://tanvrit-backups/",
    ),
) { ok -> println("scheduled=$ok") }