From intent to cluster
State your team size, budget, and workload type in plain language — Tashkil translates that intent into a fully provisioned, optimized cluster in minutes.
Compose HPC & AI Clusters in Minutes
Tashkil turns plain-language workload needs into fully provisioned, optimized clusters — across bare metal, cloud, and neoclouds — no dedicated DevOps team required.
Tashkil
Cluster Composer
Agent session · From intent to running cluster
AR
Drain node-04 for maintenance and requeue its jobs.
Thought for 3s · 2 tool calls
Drained node-04 and requeued its jobs onto spare capacity — no preemptions.
Cordon node-04done
Migrate 12 jobsdone
Health test gpu-3queued
How's spot pricing on H100s this week?
Thought for 4s
Down 8% week-over-week. us-east-2 has the deepest pool if you need sustained capacity.
Describe the cluster you need…
13b-train
Without Tashkil
With TashkilPowering research leaders
And 10,000+ more
Total flexibility across bare metal, cloud, and neoclouds
One platform that composes, optimizes, and heals every cluster
Tashkil collapses your entire infrastructure stack into a single conversational interface — built to provision, optimize, and self-heal without the orchestration tax.
Benchmark-driven hardware
Tashkil selects the most cost-effective hardware for each job from real-world performance benchmarks, not instance labels — so budgets go further.
The challenge
A dedicated cluster gives your team an edge, but standing one up yourself comes with tradeoffs
- The DevOps Bottleneck
- Weeks disappear into Kubernetes, SLURM, and cloud networking before a single job ever runs.
- Mispriced Compute
- Instance labels hide real performance, so teams overpay for hardware that doesn't match their workload.
Real reviews from real customers
The flexibility is really what made the difference. Our workloads shift weekly — I describe what the team needs and in two sentences Tashkil already has a cluster composed. That is a real advantage when you are moving quickly.
Olivier Reinaud
Co-founder at NetZero
Real outcomes from teams running Tashkil
From faster provisioning to a leaner stack, research teams skip the infrastructure grind the moment Tashkil goes live — with quotas and isolated environments from day one.
97%
Reduction in cluster setup time
6
Tools replaced by one platform
10x
Faster from intent to first job
composed every training cluster across bare metal, cloud, and neoclouds — giving each team production-ready compute on demand.
Spotify replaced six disconnected tools with Tashkil, wiring provisioning, scheduling, and node health into a single orchestrated pipeline — and went from intent to first job 10× faster.
Read the case studyOne platform, priced to scale with your team
Every plan runs the full Tashkil orchestration engine. Add capacity, automation, and governance as your program grows — billed per user, with no per-node surprises.
Starter
For small teams composing their first cluster.
$12/ user / mo
billed monthly per user
- Up to 10 seats
- Conversational cluster provisioning
- Cloud & bare-metal deploys
- 7-day job history
- Community support
Team
Most popularOrchestrated compute for fast-moving teams.
$28/ user / mo
billed monthly per user
- Everything in Starter
- Benchmark-driven hardware selection
- Automatic GPU quarantine & requeue
- 90-day job history
- SSO & SCIM provisioning
- Priority support
Scale
Governed compute for sovereign AI labs.
$48/ user / mo
billed monthly per user
- Everything in Team
- Team quotas & isolated environments
- SAML / OIDC SSO
- Immutable audit trail
- 1-year job history
- Custom domain & SLAs
Need a tailored deployment?
Talk to our team about Enterprise SLAs, on-prem and air-gapped installs, and custom data residency.
Frequently asked questions
Everything you need to know about deploying Tashkil. Can't find an answer? Our infrastructure team is one message away.
Tashkil is an agentic infrastructure platform that composes production-ready HPC and AI clusters in minutes. Describe your workload in plain language — team size, compute budget, workload type — and Tashkil translates that intent into a fully provisioned, optimized cluster, no Kubernetes or SLURM wrestling required.
Compose your next cluster in minutes, not weeks
Unify provisioning, scheduling, and governance on one platform — and let your team focus on research while Tashkil handles the orchestration.