@streamdown/code@2.0.0
asheshgoplani/agent-deck@streamdown/code@2.0.0Sep 30, 2026by github-actions[bot]
Upgrading?
AI Summary
@streamdown/code 2.0.0 is a major release that upgrades Shiki to v4 and requires Node.js 20 or newer. It improves streaming code-block highlighting by tokenizing only new lines and bounding the highlight cache. Cache lookups now compare full code and avoid caching transient streaming results.
Key Highlights
- Upgrades Shiki to v4 and requires Node.js 20+
- Tokenizes only new lines while a code block streams
- Bounds the highlight cache with an LRU limit of 5,000 results and size-based eviction
- Avoids caching transient streaming results and compares full code on lookup
Breaking Changes
- Upgrade to Shiki v4
- Node.js 20 or newer is required when using @streamdown/code
New Features
- Incremental tokenization of newly streamed lines
- Bounded LRU cache for finished code-block highlights
- Full-code cache lookup to avoid incorrect token reuse
Full Release Notes
### Major Changes - db6fefc: Upgrade Shiki to v4. Node.js 20 or newer is now required when using `@streamdown/code`. ### Patch Changes - 0ea9e64: Tokenize only the new lines of a code block while it streams instead of re-tokenizing the whole block on every update. - 4ec5639: Bound the code highlight cache. Results of streaming code blocks are no longer cached (each is superseded by the next update), and finished code blocks are kept in a least-recently-used cache bounded by their size (tokens, lines and characters) and by a limit of 5,000 results. Lookups compare the full code, so blocks that differ only in the middle no longer share tokens. A code block also ignores highlight results for code it has already moved past.
Related Repositories
Get up and running with Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.
A fast reverse proxy to help you expose a local server behind a NAT or firewall to the internet.
RAGFlow is a leading open-source Retrieval-Augmented Generation (RAG) engine that fuses cutting-edge RAG with Agent capabilities to create a superior context layer for LLMs
The world’s fastest framework for building websites.
:cherry_blossom: A command-line fuzzy finder