[Draft] What makes raged unique: semantics + stacktraces + temporal graph

[lexasub]

Status: stream of thoughts, might be nonsense. Draft for discussion, not a commitment.

The code-intelligence market is crowded. I studied competitors by reading their code:

Tool	Niche they took
codegraph (57k⭐)	zero-ops: SQLite, MCP out of the box, watcher with staleness banner
CGC (4k⭐)	quality tools: dead code, cyclomatic complexity, SCIP
CodeSlicer	impact/evidence: "what a change breaks + which tests"
Graphify (100k⭐)	community detection (Leiden), god nodes
Bitloops	assertions + capturing agent reasoning, DevQL
CCE	hybrid retrieval: vector + BM25 + RRF

Our unique combination (none of them does this)

1. Semantic search over the graph — NL search (bge-m3) + FTS enriched by the graph (search by description → symbols → callers/callees). CCE is closest, but it's not graph-first. This isn't a post-market conclusion — it's the original intent: Advanced Semantic Search with NLP and Context Awareness #20 (advanced semantic search, Feb 2026) is one of the earliest issues in this repo.
2. Stacktrace mapping — stacktraces → symbols for 4 languages. Nobody has this.
3. Temporal graph — valid_from/valid_to already in the schema! "What changed in the graph between commit A and B" — nobody does this, and the foundation is already ours.
4. Library search (Multi-Index Support for Third-Party Libraries #24) — manually index third-party libs (ast-rag index-lib --name requests --path .../site-packages/requests), search scoped by --lib. Competitors don't index libs at all (codegraph skips site-packages). "Where in requests is session timeout handled" is exactly the semantic query nobody answers — and it's the Add Error Handling for Empty Files #1 real-world agent question.

Conclusion

Winner-take-all paradigm: a user installs one tool. So:

• zero-ops (SQLite + MCP) — parity, must catch up (see [Draft] Storage: embedded SQLite instead of Neo4j+Qdrant — follow-up to #32 #57)
• quality tools — parity, algorithms known (see [Draft] Quality layer for agents: dead code, complexity, impact scoring #58)
• semantics + stacktraces + temporal + libs — our moat, what decides the choice once everything else is equal

Open question: am I overrating the semantic layer? codegraph deliberately does no embeddings (100% local, zero deps). Maybe the market has chosen "graph without semantics" — and our moat is something nobody actually needs but us?

---

Relates to: #35 (cross-repo support), #24 (multi-index for third-party libs), #20 (advanced semantic search), #31 (multi-project configs)