Forge CLI
Automated GPU kernel optimization that turns PyTorch models into drop-in CUDA/Triton kernels.
Forge delivers verified, dramatic performance gains that torch.compile can't match — up to 10× throughput on production models. But it's enterprise-only and locked to datacenter GPUs, so indie developers should stick with the Free/Pro tiers or other tools. If you run LLMs at scale on H100s, it's likely worth the investment.
Verified 4d ago · liveness 80/100 · cite: rightaichoice.com/tools/forge-cli
- ML teams optimizing inference for production at scale on H100/A100 clusters
- Infrastructure engineers maximizing GPU utilization and reducing costs
- Enterprises deploying large language models seeking 3–10× speedups
- Teams needing verified correctness alongside performance gains
- Developers needing a free or low-cost self-serve option (enterprise-only)
- Projects running on consumer/RTX GPUs (only datacenter supported)
- Teams wanting a GUI or no-code optimization (CLI-only)
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Forge if you aren't running production LLM workloads on datacenter GPUs, if you need a self-serve tool, if you're on a consumer/RTX GPU, or if you can't justify a custom sales engagement.
The Free tier of the RightNow editor only includes 1 Forge credit per month — enough for one small kernel, so you'll likely need to pay for more.
Forge is expensive and enterprise-only, but it's a fraction of the cost of the GPU savings it delivers. If you're on a budget, torch.compile is free, but Forge can save $18k/mo on GPU spend at scale.
In short
Forge CLI — Automated GPU kernel optimization that turns PyTorch models into drop-in CUDA/Triton kernels. Best for ML teams optimizing inference for production at scale on H100/A100 clusters, Infrastructure engineers maximizing GPU utilization and reducing costs, Enterprises deploying large language models seeking 3–10× speedups. Free to start; paid plans from $20/mo.
What's new in Forge CLI
Checked 2 days agoAcross the latest 4 updates: 2 feature updates, 1 launch and 1 news mention.
Forge CLI 1.0.0: Custom agents, skills, and more languages
Custom agents with skills and MCPs; added Numba, Mojo, CUDA Tile support; improved SSH and remote workflows.
LLMs Can Now Write GPU Kernels That Beat torch.compile
Multi-agent systems produce CUDA/Triton kernels with 2x to 14x speedups over torch.compile on real models like Llama-3.1-8B and Stable Diffusion.
Forge CLI Swarm Agent released
Swarm-based kernel generator with 32 agent pairs, MAP-Elites optimizer, and pattern RAG; up to 5.2x speedup over torch.compile.
PyTorch Kernel Support added
Profile, benchmark, and emulate PyTorch kernels with NCU integration, statistical timing, and 86+ GPU architecture coverage.
What people actually say about Forge CLI — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
34 mentions across 5 sources (Hacker News, YouTube, Product Hunt, GitHub, Lemmy) · researched Aug 14, 2026.
- +Delivers 3-10× speedups over torch.compile for LLM inference.
- +Automates CUDA/Triton kernel generation, saving manual tuning effort.
- +100% numerical correctness verification via tiered evaluation.
- +Supports all NVIDIA datacenter GPUs, including B200 and H100.
- +Drop-in replacement with zero code changes, simple API.
- −High cost with credit system and enterprise pricing, not for small teams.
- −Requires advanced skill level and dedicated infrastructure setup.
- −Numerical correctness verification is manual, potentially slow.
- −Confusing name overlaps with unrelated Forge projects.
- −Scarce independent community validation or benchmarks.
- • Credit system (1 credit/kernel, 1-2 for HF models) may require frequent purchases
- • Enterprise tier likely requires negotiated contract and significant upfront cost
Viability Score
How well maintained and how widely used is Forge CLI? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: September 2026
How we score →Key Features
- Automated CUDA/Triton kernel generation from PyTorch/HuggingFace models
- Swarm of 32 parallel Coder+Judge agents for concurrent generation and validation
- MAP-Elites evolutionary optimizer with 1,824 CUTLASS and Triton patterns
- Up to 5x speedup over torch.compile (Llama-3.1-8B 5.2x, Qwen2.5-7B 4.2x)
- 100% numerical correctness verification via manual review
- Automatic Tensor Core optimization (WMMA, TMA for Hopper)
- Three optimization modes: --turbo, default, --quality
- Dual output formats: Triton Python kernels and native CUDA C++
- Interactive CLI wizard and KernelBench task browser
- Session management for tracking past optimizations
- Supports HuggingFace model IDs, KernelBench tasks (250+), and custom PyTorch files
- Credit system: 1 credit per kernel, 1-2 for HuggingFace models
- Drop-in replacement: same API, zero code changes
- Kernel support for CUDA, Triton, Mojo, PyTorch, Numba (v1.0.0)
- Integrates with RightNow Code Editor and GPU emulator
About Forge CLI
Forge CLI, from RightNow AI, is an automated GPU kernel optimization engine that turns any PyTorch or HuggingFace model into a drop-in replacement CUDA or Triton kernel tuned specifically for your NVIDIA datacenter GPU. It's designed for ML teams and infrastructure engineers who need to close the gap between what they pay for and what their GPUs actually deliver. Most GPUs are only about 16% utilized out of the box; Forge pushes that to roughly 88%, delivering 3.0× faster inference, $18k/mo in GPU savings, and a 67% reduction in power — all with 100% numerical correctness verified manually. The optimization pipeline is fully automated. A swarm of 32 parallel Coder+Judge agent pairs uses MAP-Elites evolutionary search across 1,824 CUTLASS and Triton patterns to find the best kernel for your hardware. You get results in under an hour, using the same API with zero code changes. The latest release, RightNow AI 1.0.0, expands kernel support to include CUDA, Triton, Mojo, PyTorch, and Numba, covering more model types than ever before. Forge supports datacenter GPUs including B200, H200, H100, A100, and L40S. Real results are striking. On Qwen3-235B, Forge cut token generation latency by 7.6×, boosted throughput from 312 to 3,180 tokens per second (a 10× jump), and reduced cost per 1M tokens by 90%. Every optimized model is manually verified for correctness, and the output is a drop-in replacement — same API, no code changes. Forge is part of the RightNow ecosystem and comes with enterprise-grade infrastructure: dedicated compute, on-premise deployment, custom SLA and support, and NDA & IP protection. Pricing is custom and requires contacting sales, though a free demo optimizes one model with no credit card required. Unlike open-source frameworks like torch.compile(max-autotune) or manual Triton coding, Forge automates the entire optimization pipeline and verifies correctness — the direct path to maximizing GPU fleet performance if your budget and infrastructure can
Behind the Verdict
Forge is the kind of tool that makes a CFO smile. If you're running large language models on H100 or A100 clusters and your GPU utilization hovers around 16%, the math is simple: Forge can push that to 88%, cut your inference costs by up to 90%, and pay for itself in weeks. The Qwen3-235B case study — 7.6× faster latency, 10× throughput, cost per token down from $4.30 to $0.43 — is the kind of number that justifies a sales call. The catch is access. Forge is enterprise-only, priced as 'Custom' via sales. There's no self-serve tier. You need a datacenter GPU — B200, H200, H100, A100, L40S — so consumer RTX cards are out. And it's CLI-only, which is fine for infra engineers but a non-starter for teams that want a GUI. Compared to torch.compile(max-autotune), Forge wins on raw performance — up to 5× faster on Llama-3.1-8B, and it verifies numerical correctness so you don't have to trust the optimization blindly. Manual Triton coding can match or beat it in the hands of an expert, but most teams don't have that expertise on staff. Forge automates the whole pipeline, trading a custom contract for your team's time. Where it bites: the patience for a sales cycle. If you need results today, the demo optimizes one model for free — but the full deployment requires negotiation. Smaller teams or individual devs should look at the Free/Pro tiers of the RightNow Code Editor, which include profiling and GPU emulation but not Forge's kernel generation. The Pro tier is $20/mo and includes 1 Forge credit/mo, which is barely a taste — enough for a single kernel, then you're back to sales. In practice, Forge shines for enterprises with dedicated GPU fleets, where the $18k/mo savings on a single model dwarfs the cost of a contract. For everyone else, the free demo is a smart way to
Researching Forge CLI? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas Forge CLI actually fits — and what changes day-one when you adopt it.
You're running a Llama-3.1-8B inference service on H100 and want to cut latency and cost.
Outcome: You run `forge --model llama-3.1-8b --gpu H100 --baseline torch.compile(max_autotune)` and get a drop-in Triton kernel in under an hour that delivers 5.2× faster inference, with verified correctness.
You manage a fleet of A100s and are asked to reduce GPU spend.
Outcome: You use Forge to optimize your heaviest models, pushing utilization from ~16% to ~88%, saving thousands monthly, and the drop-in kernels require zero code changes.
You need to deploy Qwen3-235B at scale and want to serve more concurrent users without buying more GPUs.
Outcome: Forge optimizes the MoE attention layers, cutting latency 7.6× and boosting throughput 10×, reducing cost per token by 90% on the same hardware.
Use Cases
- Optimize Llama-3.1-8B inference on H100 GPUs for 5.2× speedup over torch.compile
- Reduce GPU costs by 67% through kernel fusion and memory optimization
- Generate production-ready Triton or CUDA kernels for any HuggingFace model in under an hour
- Benchmark and compare kernel candidates automatically with the swarm agent system
- Deploy drop-in replacement kernels with verified numerical correctness across your inference stack
- Eliminate manual low-level optimization work for custom PyTorch operations
Models Under the Hood
as of 2026-08-26
Limitations
- Forge supports NVIDIA datacenter GPUs including B200, H200, H100, L40S, A100, and more.
- Full capabilities are available through a custom sales engagement, and the Free tier of the parent RightNow editor grants only 1 Forge credit per month.
- Generated kernels are drop-in replacements with same API and zero code changes.
- Forge 0.1.0 introduced a Swarm agent with up to 5x speedup and a 97.6% correctness rate.
as of 2026-08-21
Verification history
We have re-verified Forge CLI 7 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
Showing the 6 most recent of 7 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Plans compared
For each published Forge CLI tier: who it actually fits, and what it adds vs. the previous tier. Cross-reference the cost calculator above for projected annual outlay.
Free
$0/mo
Ideal for
Solo developers or students exploring GPU kernel development with a single GPU and limited needs, who want profiling and benchmarking at no cost.
What this tier adds
Free entry point with unlimited profiling and benchmarking, 1 Forge credit per month, and limited autocomplete.
Pro
$20/mo
Ideal for
Professional teams or solo developers who need GPU emulation, multi-GPU comparison, and natural language profiling, at $20/mo.
What this tier adds
Adds GPU emulator access (50+ GPUs), multi-GPU comparison (6 max), natural language profiling, 1000 AI agent credits, and unlimited autocomplete.
Enterprise
Custom
Where the pricing makes sense
The company stage and team size where Forge CLI's pricing actually pencils out — and where peers do it cheaper.
Forge is expensive and enterprise-only, but it's a fraction of the cost of the GPU savings it delivers. If you're on a budget, torch.compile is free, but Forge can save $18k/mo on GPU spend at scale.
Setup time & first value
How long it actually takes to get something useful out of Forge CLI — broken out by persona, not the marketing-page minute.
Forge CLI installs via npm/npx in minutes. First optimization on a KernelBench task takes ~2 minutes with --turbo mode; full quality runs take under an hour. HuggingFace models may take longer.
Switching to or from Forge CLI
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From torch.compile(max-autotune): Swap the compile call with Forge and use the generated kernel as a drop-in replacement — no code changes needed.
Integrations
Resources & Guides
- Documentationrightnowai.co
Docs · Forge CLI
Full product docs from rightnowai.co
- Documentationrightnowai.co
Installation · Forge CLI
Full product docs from rightnowai.co
- Documentationrightnowai.co
Core Features · Forge CLI
Full product docs from rightnowai.co
- Documentationrightnowai.co
Advanced Features · Forge CLI
Full product docs from rightnowai.co
- Resourcerightnowai.co
Forge · Forge CLI
Helpful link from rightnowai.co
Tutorials & Learning
Official links
Tools that pair well with Forge CLI
Common stack mates teams adopt alongside Forge CLI, with the specific reason each pairing earns its keep.
Featured Head-to-Head Comparisons
Forge Cli vs Spider Cloud
Spider Cloud and Forge CLI serve completely different needs: Spider Cloud is a web data extraction API for AI agents, while Forge CLI is a GPU kernel optimizer for PyTorch models. If you need real-time web data for RAG or LLM context, Spider Cloud's freemium model and browser AI commands are the right choice. If you're an ML engineer maximizing inference speed on datacenter GPUs, Forge CLI's automated kernel generation can deliver 2–5x speedups over torch.compile, but requires contacting sales for pricing.
Forge Cli vs Temporal Ai
If you need to orchestrate reliable, fault-tolerant AI agents or microservices, Temporal AI is your pick. If your goal is maximum GPU inference speed for production models, Forge CLI delivers up to 5× faster kernels. Choose based on your bottleneck: workflow reliability vs. raw performance.
Forge Cli vs Voyage Ai
Choose Voyage AI if your priority is high-accuracy retrieval in regulated RAG workflows with long-context, domain-specific embeddings — its low-dimensional vectors and 32K token support cut storage costs and improve search. Choose Forge CLI if you need to maximize GPU inference performance for large models on datacenter hardware; recent updates show it can beat torch.compile by up to 14x with verified correctness, though it requires contacting sales for pricing and only supports enterprise GPUs.
Alternatives to Forge CLI
View allFrequently Asked Questions
Used Forge CLI? Help shape our editorial sentiment research.