Products
Fast Apply (Apply) — File editing model that merges partial code updates into existing files at 10,500 tok/s with 98% accuracy. Uses the<instruction>/<code>/<update> message format.
WarpGrep — Code search subagent that runs in a separate context window. Explores repositories using built-in tools (grep_search, read, list_directory, glob, finish). Returns matching code snippets in ~6 seconds.
Compact (Compactor) — Context compression model that removes irrelevant lines from chat history and code at 33,000 tok/s. Every surviving line is byte-for-byte identical to the original input.
Router — Prompt complexity classifier that returns a model recommendation in ~430ms. Does not generate completions itself. Routes to the optimal model (e.g., claude-haiku for simple, claude-sonnet for complex).
Model IDs
| Model | ID | Purpose |
|---|---|---|
| Apply (fast) | morph-v3-fast | Default file editing, highest speed |
| Apply (large) | morph-v3-large | Complex edits requiring more reasoning |
| Apply (auto) | auto | Router selects fast vs large automatically |
| WarpGrep | morph-warp-grep-v1 | Codebase search (local and GitHub) |
| Compact | morph-compactor | Context compression |
| Embedding | morph-embedding-v4 | Code and text embeddings |
| Rerank | morph-rerank-v4 | Search result reranking |
| Router | morph-routers | Prompt complexity classification |
Message Format Tags
<instruction> — XML tag in Apply messages describing what the edit does. Including it raises accuracy from 92% to 98%.
<code> — XML tag containing the original file content to be edited.
<update> — XML tag containing the partial edit snippet with // ... existing code ... markers for unchanged regions.
// ... existing code ... — Marker placed in <update> snippets to indicate regions that should remain unchanged. Required for Apply to correctly merge partial edits.
<repo_structure> — XML block in WarpGrep messages describing the repository directory layout. Required for local codebase search.
<keepContext> — XML tag wrapping sections of input that Compact should never remove, regardless of relevance scoring.
API Concepts
Base URL —https://api.morphllm.com/v1. All endpoints are OpenAI-compatible and work with any OpenAI SDK.
Bearer token — Authentication method for all Morph API endpoints. Obtained from the dashboard.
query parameter (Compact) — Tells Compact what information matters for the next LLM call. Without it, the model infers relevance from the last user message.
code_context — Parameter in the edit_file tool definition containing the original file content to be edited.
search_context — Parameter in the codebase_search tool definition containing the repository structure for WarpGrep queries.