Skip to content

roadmap.md

Latest commit

 

History

History
114 lines (92 loc) · 4.78 KB

roadmap.md

File metadata and controls

114 lines (92 loc) · 4.78 KB
Raw

Roadmap

See PLANNING.md for the full detailed roadmap and brainstorming document.

✅ Completed / Shipped

Chunking & Indexing

  • AST-based code chunking for 17 languages
  • Regex/document chunking for Markdown, Razor, .sln, LaTeX
  • Document text extraction for PDF, DOCX, DOC, Excel
  • Image indexing via vision LLM (descriptions stored as searchable vector chunks)
  • Line-based fallback chunking
  • Pluggable chunkers via config
  • Per-language chunking config (chunking.nodeTypes)
  • Incremental indexing (file-hash-based, manifest-backed)
  • File watching and background re-indexing
  • Enhanced chunk descriptions with relative paths and line numbers

Embedding & Storage

  • Embedding providers: Ollama, OpenAI, Cohere
  • Proxy-aware embedding transport with raw socket localhost bypass
  • Dimension probing at startup (persisted to config to reduce startup time)
  • LanceDB vector storage with memory:// test mode
  • Batch embedding with progress reporting
  • Auto-detection of LanceDB schema for seamless upgrades
  • Persistent image description cache (reused across sessions)

Retrieval

  • Vector search pipeline
  • Hybrid search (TF×IDF keyword + vector fusion)
  • Session-level retrieval cache
  • Auto-context injection on chat.message
  • Configurable auto-inject settings
  • Context window optimization (per-file chunk limits, adjacent merge, Jaccard dedup)

OpenCode Plugin

  • search_semantic tool
  • chat.message hook with file suggestions and auto-injection
  • RAG-backed read override tool
  • TUI settings panel with model picker dropdowns
  • OpenCode v1.17.0 compatible PluginModule export
  • Background auto-indexing with watcher status
  • API key auto-resolution from OpenCode provider config
  • Documentation mode (/doc slash command, per-subdirectory progress tracking)
  • Wiki mode (/wiki slash command, AI-maintained knowledge wiki at .opencode/wiki/)
  • Hotkey-activated context injection (Ctrl+Enter / Ctrl+Alt+Enter)

CLI & Distribution

  • Full CLI: init, index, query, clear, status, list, show, dump, describe-image, ui, mcp, setup
  • Evaluation CLI: eval:sessions, eval:analyze, eval:compare
  • init command lifecycle with plugin generation, gitignore, npm install
  • AGENTS.md creation/merge via sentinel markers
  • MCP server (opencode-rag mcp) exposing semantic search tools over stdio
  • Web dashboard UI (opencode-rag ui)
  • Install scripts (.sh / .ps1) with uninstall support
  • Release automation script
  • Published npm package: opencode-rag-plugin

Configuration & Quality

  • JSON config with deep-merged partial overrides
  • Runtime overrides system for live TUI changes
  • Configurable file logging
  • Manifest schema versioning with corruption detection
  • Persisted embedding vector dimension (probed once, cached in config)
  • Path traversal protection in file resolution
  • 589+ automated tests

Short Term

Feature Description
LLM-based re-ranking Cross-encoder or lightweight model after vector search
Query rewriting Multi-variant expansion for ambiguous queries
Persistent query cache Disk-based cache so repeated queries across restarts are instant
Concurrent chunking Parallel file scanning/chunking for large repos

Mid Term

Feature Description
Cross-file relationship graph Import/call graph for dependency-aware search
Multi-repo search Index and search across multiple workspaces
IDE context awareness Use current file, cursor position for relevance boosting
Prompt customization Customize how retrieved context is formatted
Persistent session memory Retain coding patterns and decisions across sessions
Auto-generated codebase summaries LLM directory-level summaries from indexed chunks
Chunk quality heuristics Score chunks for size, coherence, boundary quality

Long Term

Feature Description
Code execution-aware retrieval Run code to understand its behavior for better retrieval
Semantic refactoring assistant Code transformations based on natural language
Agent-based code navigation Autonomous exploration of codebase structure
Multimodal support Diagrams, API specs, JSON schemas, YAML configs
Access control Per-folder permissions, sensitive file exclusion
Index export/import Serialize index for CI/CD, team sharing, backup/restore
Performance benchmark suite Measure index time, query latency, memory usage
Memory & storage optimization Quantized embeddings, pruning, garbage collection

Key Next Steps

  1. LLM-based re-ranking for retrieval precision
  2. Code graph integration for structural code understanding
  3. Query rewriting for ambiguous query expansion
  4. Persistent session memory across coding sessions
  5. Concurrent chunking for faster indexing of large repos