Morph: Apply AI edits to files FAST

Fast Apply merges an AI edit into a file at 10,500 tokens per second, with 98% accuracy. The speed comes from a small model trained end to end for one job, served on a custom inference engine with hand-written GPU kernels, not a general-purpose endpoint. The result is sub-second latency on typical files, which matters when an agent edits code many times in a session. For higher-accuracy batch work, morph-v3-large trades speed for precision.

The model works from a lazy edit, which is just the changed lines wrapped in // ... existing code ... markers. The agent sends only what it changes, and Fast Apply returns the complete merged file. For LLM code edits that land as fragments, that means the merge happens in one call instead of a client-side patch routine. This cuts token use by roughly 40% against a full-file rewrite, because nobody pays for the unchanged lines twice. The one catch is that missing markers read as deletions, so your agent prompt has to teach the format.

Everything runs through one OpenAI-compatible endpoint at https://api.morphllm.com/v1. Existing OpenAI code only needs a new base URL and model string. Morph ships a TypeScript SDK, @morphllm/morphsdk, and Python users run the standard OpenAI client. The SDK also has ready-made edit tools for Anthropic, OpenAI and the Vercel AI SDK, so you can drop Fast Apply behind an edit_file tool call without writing the glue yourself.

Morph ships an MCP server that adds an edit tool and codebase search to your editor. A single command, npx -y @morphllm/morph-setup, auto-detects Claude Code, Cursor, Codex and VS Code and configures them all. That path is for people who want the speed-up without building an agent from scratch. Per-client setup guides and prompt templates cover the rest.

The free tier covers 200 requests a month, which is enough to test Fast Apply on real work before you pay. Past that, pricing is per token with no per-seat fees: morph-v3-fast runs $0.80 per million input tokens and $1.20 per million output tokens. Subscription plans start at $20 a month for 2M credits and scale to $400 a month for 80M credits on the team tier.

Morph offers self-hosted and air-gapped deployments for teams that cannot send code to a hosted API. Enterprise Apply adds custom model configurations, and the business side covers SOC 2 Type II, GDPR and data residency. For companies with a security review to pass before any code leaves the building, that path is the one that matters.

Fast Apply handles text code in any language, whether that is TypeScript, Python, SQL or a config file, because it works on the file's text rather than a language-specific parser. It is built for single files and multi-hunk edits within them. Binary files, minified JavaScript and non-text content are out of scope, and single-character edits are cheaper to do client-side.