Run any job on
any machine.
Install the agent on your machine and join a shared compute network. Your idle CPU runs jobs for others — you earn credits and spend them running jobs on everyone else's machines. The more users, the more power every user gets.
$ compute job submit \ --type DOCKER \ --image ghcr.io/myorg/runner:latest \ --command "pytest tests/ -x" \ --label gpu=A100 --timeout 600 ● Submitted job_01jt4k8n2q · queued → assigned → running ● Streaming logs from node gpu-a100-fra-01 PASSED tests/test_model.py::test_inference ✓ SUCCESS 47.3s · 0.24 credits
A full portal, on every device
The same portal runs natively on macOS, Windows, Linux, Android, iOS, and in the browser — built with Kotlin Multiplatform Compose so the UI is identical everywhere.
Everything a compute platform needs
Built for teams who need real job orchestration — not cron wrapped in bash.
Unified job model
COMMAND, DOCKER, GITHUB_ACTION, AI_TOOL — all under one submit API. The scheduler picks the best available node automatically.
DAG dependencies
Chain jobs into one-level DAGs. Downstream jobs block until their dependencies succeed. Failed upstreams auto-cancel dependents.
GPU marketplace
Bid on spare GPU capacity from community hosts — Vast.ai style. A100, H100, L40S. Per-second billing, cancel any time.
Multi-region HA
Raft-based control plane with automatic leader election. Deploy standby replicas in EU, US, and AP — jobs survive region failures.
Cron schedules
Recurring jobs with jitter, backoff, and pause/resume. Run-now override from the portal, CLI, or SDK without touching the schedule.
Template gallery
Fork-and-run: Android builds, iOS XCTest, Llama inference, Blender renders, ffmpeg, pytest, k6 load tests. One click to customise.
Agent auto-update
Nodes self-update to new agent versions without manual SSH. Version pinning, staged rollout, and rollback on health check failure.
API keys & audit
Per-project API keys with label-based scoping. Every submit, cancel, and retry is logged and surfaced in the portal audit trail.
Tanvrit-native
First-class integration with Tanvrit AI inference and Tanvrit Market campaign pipelines. Shared credits, shared identity.
Like BitTorrent — but for compute.
Every machine you add is both a contributor and a consumer. 100 users means every user has the combined power of 100 machines at their fingertips — for free, as long as the network stays balanced.
- ✗ Pay for runners even when idle
- ✗ Queue spikes on push-heavy days
- ✗ One team's idle = wasted money
- ✗ Scale = provision more VMs (cost ↑)
- ✗ Your build waits behind everyone else
- ✓ Idle machines earn credits automatically
- ✓ Queue drains across the whole network
- ✓ One team's idle = another team's free build
- ✓ Scale = more users joining (cost stays $0)
- ✓ Labels let you target mac/gpu/xcode nodes
Contribute more than you consume → your balance stays positive → compute stays free. Heavy users top up with pay-as-you-go credits ($0.0001/sec).
| Platform | Submit jobs | Contribute compute | Notes |
|---|---|---|---|
| macOS agent | ✓ | ✓ | Full — shell, Docker, GitHub Actions |
| Linux agent | ✓ | ✓ | Full — shell, Docker, GitHub Actions |
| Windows agent | ✓ | ~ | Shell via cmd; Docker requires Docker Desktop |
| macOS / Windows / Linux app | ✓ | — | Portal only — submit & monitor jobs |
| Android app | ✓ | — | Portal only — OS kills background processes |
| iOS app | ✓ | — | Portal only — sandboxed, no background execution |
| Web (browser) | ✓ | — | Portal only — no subprocess execution |
The agent (macOS / Linux / Windows) is what joins the shared pool. All portal apps can submit to it from any device.
Native apps for every platform
The Tanvrit Compute portal is a native app built with Kotlin Multiplatform Compose. Same codebase, pixel-perfect on each OS.
Desktop
Or use it right now in your browser → Open web portal
Up and running in minutes
No Kubernetes. No Terraform. No YAML sprawl.
Join the network
Install the agent on any machine — laptop, server, or VM. One command: it registers, starts earning credits by running others' jobs when idle, and unlocks the entire network for you.
curl -fsSL https://compute.tanvrit.com/install.sh \ | TANVRIT_API_KEY=$KEY sh # Your machine is now in the shared pool.
Submit a job
Use the SDK, CLI, or portal to submit a job. Add labels to target specific node pools, set priorities, or attach it as a downstream step in a DAG.
compute job submit \ --type COMMAND \ --command "cargo build --release" \ --label os=linux,arch=arm64
Stream results
Logs stream in real-time via SSE. The job status, stdout, and exit code appear in the portal, CLI, and SDK — with webhook notifications on completion.
compute job logs job_01jt4k8n2q --follow
One SDK across all your platforms
The Tanvrit Compute SDK is a Kotlin Multiplatform library — the same code runs on Android, iOS, JVM, and WASM. Import it once and ship jobs from wherever your app lives.
- ✓ Android, iOS, Desktop, Web — one artifact
- ✓ Coroutines-native: suspend functions, Flow
- ✓ Koin DI out of the box
- ✓ ViewModels with lifecycle awareness
val vm: ComputeJobViewModel = koinInject() val jobs by vm.allJobs.collectAsState() // Submit a job vm.submitJob( SubmitComputeJobRequest( type = "DOCKER", image = "ghcr.io/org/runner", command = "pytest tests/", gpu = GpuRequest(count = 1), ) ) { result -> snackbar.show("Submitted: ${result.jobId}") } // Stream real-time logs vm.streamLogs(jobId) { line -> logBuffer.append(line) }
Compute is free if you contribute
Your idle machine earns credits. Spend them on any node in the network. Top up only if you consume more than you give.
- ✓ Join the shared pool with 1 command
- ✓ Earn $0.0001/sec per core while running jobs
- ✓ Spend credits on any node in the network
- ✓ Net contributors never pay
- ✓ Everything in Contributor
- ✓ Access GPU nodes (A100, H100, L40S)
- ✓ Priority scheduling
- ✓ Email support
- ✓ Dedicated node pool — no sharing
- ✓ HIPAA / SOC 2
- ✓ 99.9% SLA
- ✓ Volume discounts
One command to join the network.
Install the agent, earn credits from your idle machine, and immediately access every other node in the network. No credit card. No Kubernetes.