ArgosBrain vs Continue.dev

What Continue.dev is

The baseline, stated fairly.

Continue.dev (Apache-2.0) pairs @codebase retrieval with an optional mcp/memory block for persistent cross-session memory. @codebase parses the workspace with tree-sitter .scm queries to extract top-level functions/classes, embeds chunks with all-MiniLM-L6-v2 locally, and stores them in LanceDB.

Retrieval combines embedding similarity + keyword search.

How it actually works

Technical facts.

StorageLanceDB on disk (per-workspace); mcp/memory block uses Anthropic's reference MCP memory server (JSON file).
RetrievalTop-K chunk injection on @codebase; MCP memory injected on demand in Agent mode.
CostEmbedding free (local), but retrieved chunks are injected into the prompt — pay prompt tokens per query equal to nRetrieve/nFinal × chunk size.
LimitationsIssue #7072 — @Codebase doesn't include relevant files (recall gaps); transformers.js embeddings not supported in JetBrains (#1650); MCP only usable in Agent mode; protocol lag (#8118).

Sources: Codebase docs · MCP docs

Verdict

Where each one wins.

↑ Where ArgosBrain wins

Compiler-grade semantic graph vs tree-sitter chunks. Continue extracts top-level fn/class text blocks for embedding — no cross-file symbol resolution, no type info, no inheritance, no call edges. ArgosBrain's tiered pipeline (SCIP / LSP / bespoke tree-sitter drivers with semantic hooks) extracts canonical symbol IDs, containment, call graph, overrides, and module boundaries as first-class graph edges.
Symbol-level precision API (symbol_exists, resolve_member, contained_by) that Continue's chunk-level index can't provide.
Graph traversal — answering "who calls X" or "who overrides Y" is a traversal, not a cosine similarity.
$0 retrieval. Continue pays prompt tokens per query.
Recall. Community issues confirm @codebase misses relevant files at scale.

↑ Where Continue.dev wins

Inline completion + chat UX inside VS Code / JetBrains is mature.
LanceDB + local embeddings is a solid baseline for semantic chunk retrieval.

When to choose which

Honest recommendation.

Choose Continue.dev if

You want an OSS Copilot alternative
You need basic workspace RAG

Choose ArgosBrain if

Keep Continue for completion, add ArgosBrain as the MCP memory server for deterministic structural queries