Tanvrit Compute
Open portal

Getting started with Tanvrit Compute

> Install the agent, submit your first job, stream logs in the portal.

1. Install the agent on a machine you own

curl -fsSL https://compute.tanvrit.com/install.sh | sh

This drops tanvrit-agent into ~/.tanvrit/bin/ and prints instructions. On macOS you can also brew install it:

brew tap tanvrit/compute
brew install tanvrit-agent

2. Register the agent

Sign in at compute.tanvrit.com and open Settings → Agent API keys → Generate key. Copy the raw key (it is shown once).

Run the agent with the key:

TANVRIT_SERVER_URL=https://api.tanvrit.com \
TANVRIT_API_KEY=<your-key> \
tanvrit-agent

You should see the agent print:

┌──────────────────────────────────────────────────────┐
│           Tanvrit Compute Agent v0.2.0               │
├──────────────────────────────────────────────────────┤
│  Server: https://api.tanvrit.com                     │
│  Node:   <hostname>-macos                             │
│  CPU:    8                                            │
│  Memory: 16384MB                                      │
│  Labels: macos, arm64, docker, metal, gpu, xcode     │
└──────────────────────────────────────────────────────┘
✓ Registered as node: cn-<uuid>

The node now appears in the portal's Nodes tab as ONLINE.

3. Submit a job from the CLI

tanvrit-compute login
tanvrit-compute submit "echo hello from compute" --label macos

Output:

Job submitted: cj-<uuid>
Streaming logs (ctrl-C to detach):
hello from compute
✓ Completed in 42ms (exit=0)

4. Submit a job from Kotlin (SDK)

import com.tanvrit.compute.di.computeModule
import com.tanvrit.compute.engine.dsl.computeJob
import com.tanvrit.compute.feature.job.handler.ComputeJobHandler

// one-time: start Koin with the compute module
startKoin { modules(computeModule) }

// submit a job
val request = computeJob {
    shell("python train.py --epochs 10")
    gpu("cuda")
    memory(16384)
    priority(8)
    tag("experiment-42")
}
ComputeJobHandler.shared().repository  // use the repository / viewModel to watch progress

5. Schedule a recurring job

From the portal Schedules → New schedule, pick a cron expression and a template (or a raw command). The control plane dispatches the schedule every minute.

From Kotlin:

import com.tanvrit.market.compute.TanvritMarketComputeIntegration

TanvritMarketComputeIntegration.install()
TanvritMarketComputeIntegration.scheduleWeeklyCampaignRefresh(
    businessId = "biz-123",
    campaignId = "cmp-spring-sale",
    viewModel = computeScheduleViewModel,
) { schedule ->
    println("Scheduled: ${schedule?.scheduleId}")
}

6. What to do next