Revise: AI Editor Built with Agentic Coding Tools and Y.js CRDT
Technical Implementation
Revise was built over 10 months using agentic coding tools, with the developer reporting never moving faster in their development career. The architecture includes a word processor engine and rendering layer built entirely from scratch. The only third-party library used is Y.js for the CRDT (Conflict-Free Replicated Data Type) stack.
AI Features
The editor integrates multiple AI models for document processing:
- GPT-5.4 Mini
- GPT-5.4
- GPT-5.4 Pro
- Claude Haiku 4.5
- Claude Sonnet 4.6
- Claude Opus 4.6
The AI agent works side-by-side with users in the same interface, providing inline proofreading and revision suggestions. It can detect inconsistencies in documents, such as timeline mismatches or contradictory statements about revenue changes.
Document Processing Capabilities
- Import Word documents and Google Docs
- Extract content from PDFs using multi-modal LLMs
- Proofread, revise, and perfect existing documents
- Export functionality with various settings
User Customization
Revise learns user preferences over time, including:
- Pronoun preferences (e.g., they/them)
- Feedback style preferences (direct vs. fluffy praise)
- Spelling preferences (US vs. other variants)
- Formatting preferences (APA 7, Oxford comma usage)
- Tone preferences (warm and precise, avoiding exclamation points)
- Accessibility considerations
- Document structure preferences (short paragraphs, concise drafts)
Technical Writing Preferences
The system supports specific technical documentation requirements:
- Performance claims must include benchmark setup, workload shape, sample size, and baseline comparisons
- Preference for concrete API contracts with request/response payloads, pagination rules, and idempotency semantics
- Breaking changes must be called out (API changes, data backfills, index migrations, client version gates)
- Requirements as numbered acceptance criteria with edge cases, retry behavior, and expiration details
- Trust in metrics like p95s, error budgets, queue depth, and incident counts over adjectives
📖 Read the full source: HN AI Agents
👀 See Also
OmniCoder-9B fine-tune shows strong performance for agentic coding on 8GB VRAM systems
A Reddit user tested OmniCoder-9B, a fine-tune of Qwen3.5-9B on Opus traces, with OpenCode and reported 40+ tokens per second speeds using Q4_K_M GGUF quantization at 100k context length on an 8GB VRAM system.
Jan-Code-4B: A Lightweight Code-Tuned Model for Local Development
The Jan team released Jan-Code-4B, a 4B parameter code-tuned model based on Jan-v3-4B-base-instruct. It's designed as a drop-in replacement for Haiku in Claude Code, offering improved coding assistance while running locally.
Canary: AI QA Agent for Automated Testing Based on Code Changes
Canary is an AI QA agent that reads codebases, analyzes pull request diffs, and generates end-to-end tests for affected user workflows. It connects to preview environments, runs tests, and comments results directly on PRs with recordings.
OpenClaw skill reduces accessibility tree tokens from 600K to 1.3K for ad-heavy sites
A developer built an OpenClaw skill that uses ML-based element ranking to prune accessibility trees, cutting slickdeals.com from ~598K tokens to ~1.3K tokens by keeping only the top ~50 actionable elements.