Public Roadmap
Travsr ships in themed phases, each a set of sprints with a clear exit criterion. This page reflects what is shipped today and what is next.
Phases
| Phase | Theme | Status |
|---|---|---|
| Phase 1 | MVP: CLI, SQLite, BFS, MCP stdio | Shipped |
| Phase 2 | Stability: VS Code extension, daemon + watcher correctness | Shipped |
| Phase 3 | Full retrieval stack: PPR, PCST, k-core, 0-1 knapsack | Shipped |
| Phase 4 | Language coverage: 16 languages via the Phase B plugin host | Shipped |
| Phase 5 | Cloud Preview: hosted MCP SSE, cloud indexing, team repos | In progress |
| Phase 6 | Storage backend for very large graphs, PPR/PCST at scale | Planned |
What is shipped today
- 16 languages parsed structurally by Tree-sitter (Phase A): TypeScript/JavaScript, Rust, Python, Go, Java, Kotlin, Scala, C#, Ruby, PHP, C, C++, Swift, Objective-C, and Dart.
- Semantic call edges (Phase B). Native and automatic for TypeScript/JavaScript, Rust, and Python; installable per language via
travsr lang install <lang>for the rest, each running in a sandboxed subprocess. - Phase B plugin host. A two-transport architecture (in-process native plus sandboxed sidecar) with a per-language trust model across three sandbox tiers.
- Full retrieval stack. BFS, Personalized PageRank, Prize-Collecting Steiner Tree, k-core, BM25 lexical ranking, an optional cross-encoder reranker, and a 0-1 knapsack token-budget enforcer.
- 26 MCP tools in a single-repo stdio session, 17 of them over SSE, including
get_context,get_execution_path,get_graph_json,get_snippets, and the read-only observability toolsget_index_status,get_graph_health, andget_daemon_logs. - VS Code extension with a live Cytoscape graph panel plus Languages, Synonyms, Repos, and Graph Stats panels.
- Multi-repo global registry: every
travsr initregisters a repo; one--globalMCP server serves them all. - SSE server (
travsr serve) for cloud and team deployments.
What is next
- Cloud Preview (Phase 5). Hosted MCP SSE endpoint, cloud-side indexing, and team repository support.
- Storage for very large graphs (Phase 6). SQLite + WAL is the only backend today and has been sufficient for every target workload, so this is a ceiling question rather than a scheduled swap. Kùzu was once the planned answer and was dropped; if a second backend is ever needed, it will be chosen and designed fresh.
Feedback
Open a GitHub Issue to suggest a feature or report a bug.