RightNow AI is a YC-backed GPU research lab and software company. Its flagship product is RightNow Agent, the fastest coding agent harness in the world. It also builds the RightNow editor and Forge.

What is RightNow Agent?

RightNow Agent is the fastest coding agent harness in the world, a Rust terminal CLI built for one thing: speed. On the measured Windows path, a warm daemon opens a session in 105 to 115 ms.

Which NVIDIA GPUs does the RightNow editor support?

The RightNow editor supports all NVIDIA GPUs with CUDA Toolkit 11.0-12.5, including GeForce RTX 40/30/20 series, GTX 16/10 series, Quadro RTX, Tesla, A100, and H100.

How much does RightNow AI cost?

RightNow AI is free to use with unlimited profiling and benchmarking. RightNow Pro costs $20 per month and adds GPU emulator access (50+ GPUs), multi-GPU comparison, and 1,000 AI credits per month.

What CUDA development workflow does the RightNow editor support?

The RightNow editor supports CUDA development workflows that combine editing, profiling, emulation, remote GPU execution, and benchmarked performance analysis.

Can I use the RightNow editor on macOS?

Yes, the RightNow editor is available on macOS (Apple Silicon and Intel). Mac users can use remote GPUs for free or our built-in GPU emulator for CUDA profiling.

Security

RightNow Agent runs tools locally and sends model inference to a configured provider. The shipped profile sends prompts and relevant tool results to RunInfra at https://api.runinfra.ai/v1. It is not an air-gapped product.

Credentials

The shipped model credential is read from RIGHTNOW_API_KEY.
A custom Cerebras entry can name CEREBRAS_API_KEY.
Keep credential values in the environment or an approved secret mechanism, not in committed config, documentation, receipts, or shared logs.

Receipt projections omit header and query values, but the full effective sampling config contributes to a hash. Review every diagnostic artifact before sharing it.

Network boundary

Model requests go to the selected provider.
Web search and web fetch contact public sites when called.
MCP servers, hooks, proxies, and custom providers keep their own command, network, and credential behavior.
remote_fetch = false disables optional model-catalog fetching only.
telemetry = false does not disable model or tool traffic and does not remove the local unified log.
The updater makes no installation decision because the product channel and artifact verifier are disabled.

Permissions are decision gates

--always-approve, --yolo, and --dangerously-skip-permissions auto-approve tool execution. Do not use them for untrusted work.

On Windows, permission prompts are application-level decisions, not kernel-enforced filesystem or network isolation. Use an external containment boundary when stronger isolation is required.

Untrusted input and integrations

Treat repository text, command output, fetched pages, MCP output, hooks, and migrated rules as untrusted input. Review commands, paths, destinations, URLs, headers, and environment values before approval. The shipped template disables machine-wide Claude and Cursor skill, rule, and agent imports. MCP migration is opt-in.

Report vulnerabilities privately

Do not publish credentials, customer data, or exploit details in a public issue. Include the affected version and operating system, impact, reproduction steps, redacted configuration, and disclosure status when possible.

The repository security policy does not yet specify an approved private reporting channel. That placeholder must be resolved before a release directs reporters to it.