Not a search engine for code. A structural understanding layer.

Deep Semantic Search

Understand intent across languages with vector embeddings. Powered by Azure OpenAI text-embedding-3-large, batched and parallelised at index time.

Codebase Knowledge Graph

Connect commits, work items, PR reviews, code reviews, call graphs, import graphs, and type hierarchies into a single, queryable knowledge graph with temporal scoring.

Cross-Repository Context

Materialised contract maps surface producers and consumers across every indexed repo. One question, every caller. No per-repo blind spots.

Call Graph & Impact Analysis

Static call-site extraction with multi-pass name resolution across 14 languages. Trace forward execution from any HTTP route or entry point. Reverse BFS finds all upstream callers with log-scaled risk scoring. Know the blast radius before you touch a line.

Native MCP Integration

34 MCP tools plus a transparent, secure API proxy for your AI assistants. search_code, contract_map, trace_flow, structural_impact, find_impact_graph, detect_communities, and more.

Architectural Community Detection

Union-Find clustering over call graph edges automatically discovers architectural modules. Labels by dominant directory, identifies entry points, and generates architecture docs without any LLM cost.

Module Briefings & Scope Packs

Per-community architecture overviews with entry points, and per-directory API surface bundles with import graphs. Compiled at index time.

Symbol Contexts & Contract Maps

Per-symbol bundles with callers, callees, work items, review context, and commits. Plus materialized cross-repo contract topology. Zero LLM cost at query time.

Multi-pod safe

Workers heartbeat into the queue. If a pod disappears mid-job, another picks the work up from where it stopped. No silent stalls, no duplicate runs.

Deploy-resilient

Jobs interrupted by a rolling deploy are auto-restarted at the front of the queue. Canary deploys ship with telemetry baselines, dev environments serialise to keep things sane.

Priority-aware queue

Priority-weighted dequeue and dynamic per-job timeouts. A 30-minute mono-repo full reindex won't starve the canary checking a five-file change.

Three modes, pick per repo or per run:

• FullRe-parse, re-embed, re-compile artifacts from scratch.
• IncrementalOnly touched files since the last indexed commit.
• RecompileKeep the index, rebuild the artifacts. Fast.

Add your Engram MCP configuration to an AI tool.

{
  "mcpServers": {
    "engram": {
      "url": "https://..."
    }
  }
}