Skip to main content

REPOSITORY INTELLIGENCE

Understand Any codebase.

AST PARSING DEPENDENCY GRAPHS CALL GRAPHS RAG IMPACT ANALYSIS

ARIA maps files, symbols, dependencies, call relationships, and change impact into evidence-backed engineering intelligence.

EXPLORE ARIA → VIEW SOURCE ↗
DETERMINISTIC ANALYSIS
PRESS ⌘K TO SEARCH

Structure is hidden. Until you map it.

01 · THE PREMISE NEXT — STRUCTURAL TRANSFORMATION
01 — THE PREMISE

The codebase is not a file tree.

Thousands of files hide relationships that are difficult to see, trace, and reason about.

01 — THE PREMISE NEXT → STRUCTURAL TRANSFORMATION
02 — STRUCTURAL TRANSFORMATION

The repository stops being a list.
It becomes a topology.

FILES · what is listedILLUSTRATIVE · ARIA'S OWN REPOSITORY

The repository stops being a list. It becomes a topology.

STRUCTURE → TOPOLOGYOnce relationships are resolved, position stops being alphabetical.

03 — STRUCTURAL INTELLIGENCE

Every import. Every symbol. Every caller. Every dependency.

NETWORKX · DIRECTED GRAPH
CALLERSIMPORTS
ILLUSTRATIVE · ARIA'S OWN REPOSITORY

Repository dependency graph, 7 modules shown of 1,284 indexed nodes and 3,940 edges. Selected module: services/analysis_orchestrator.py. Root orchestration hub. 48 callers, 12 imports, PageRank 94 percent. Figures are illustrative, drawn from ARIA's own repository.

  • services/analysis_orchestrator.py — Root orchestration hub, 48 callers, 12 imports, PageRank 94 percent.
  • core/ast_parser.py — Symbol extraction core, 34 callers, 4 imports, PageRank 82 percent.
  • core/dependency_graph.py — Topological engine, 29 callers, 6 imports, PageRank 76 percent.
  • backend/routers/repositories.py — HTTP entry surface, 21 callers, 8 imports, PageRank 68 percent.
  • services/github_service.py — Remote acquisition, 16 callers, 3 imports, PageRank 58 percent.
  • mcp/tools/search_tools.py — Retrieval bindings, 18 callers, 5 imports, PageRank 62 percent.
  • models/schemas.py — Contract layer, 42 callers, 1 imports, PageRank 72 percent.
SELECTED MODULESERVICE

services/analysis_orchestrator.py

Central orchestration engine managing AST detection, graph indexing, and SSE streaming pipeline.

WHYNothing runs without it — every analysis stage is sequenced here.

CALLERS
48
IMPORTS
12
PAGERANK
94%
CHANGE PROPAGATION
7 nodes directly connected
TRACE IMPACT →
SELECT ANOTHER MODULE

TOPOLOGY → BLAST RADIUSWhat is connected tells you what can be affected.

04 — CHANGE SURFACE

Know what a change touches.

Blast radius traced through the call graph: direct callers, downstream dependencies, and the API surface a change reaches.

SIMULATE A DIFF
DIFF ≠ IMPACTCALL GRAPH PROVEN
AFFECTED SURFACE
0files reached
DEPTH
4
SYMBOLS
28
HOPS
4

STATE · CHANGED SYMBOL

ILLUSTRATIVE · ARIA'S OWN REPOSITORY

  1. CHANGED SYMBOL

    ASTParser.extract_call_hierarchy()

    Signature gains an optional AST scope parameter.

  2. DIRECT CALLER

    DependencyGraph.build_call_graph()

    Consumes the node relationship tuple it returns.

  3. ORCHESTRATOR

    AnalysisOrchestrator.run_pipeline()

    Parallel worker payload structure shifts with it.

  4. ENTRY POINT

    GET /api/v1/analysis/{owner}/{repo}

    Client SSE stream response contract is affected.

A change is never isolated.

Traced through the call graph before the merge — not inferred from the diff text.

BLAST RADIUS → REPOSITORY MEMORYStructure is one dimension. Time is the other.

05 — REPOSITORY MEMORY

Churn, commit concentration and architectural hotspots read from git history — what the repository is, and how it got that way.

TEMPORAL TELEMETRY

Structure tells you how the repository is arranged. History tells you which parts of it are still moving.

COMMITS
771
MODULES
7
HOTSPOTS
2
ROOT ORCHESTRATION HUBHOTSPOT

Touched in every phase of the project, and still changing.

214 COMMITS·92% REWRITTEN

ILLUSTRATIVE · ARIA'S OWN REPOSITORY

202420252026

  1. services/analysis_orchestrator.py

    ROOT ORCHESTRATION HUB
    214 commits

  2. core/ast_parser.py

    SYMBOL EXTRACTION CORE
    168 commits

  3. core/dependency_graph.py

    TOPOLOGICAL ENGINE
    131 commits

  4. models/schemas.py

    CONTRACT LAYER
    97 commits

  5. backend/routers/repositories.py

    HTTP ENTRY SURFACE
    64 commits

  6. mcp/tools/search_tools.py

    RETRIEVAL BINDINGS
    58 commits

  7. services/github_service.py

    REMOTE ACQUISITION
    39 commits

Change history across 2024 to 2026, for 7 modules. Figures are illustrative, drawn from ARIA's own repository.

  • services/analysis_orchestrator.py — ROOT ORCHESTRATION HUB, 214 commits, 92 percent rewritten, architectural hotspot. Touched in every phase of the project, and still changing.
  • core/ast_parser.py — SYMBOL EXTRACTION CORE, 168 commits, 74 percent rewritten, architectural hotspot. Heavy early churn as language support widened, quieter since.
  • core/dependency_graph.py — TOPOLOGICAL ENGINE, 131 commits, 63 percent rewritten. Rewritten once when centrality replaced simple import counting.
  • models/schemas.py — CONTRACT LAYER, 97 commits, 41 percent rewritten. Small, constant edits — the cost of being imported everywhere.
  • backend/routers/repositories.py — HTTP ENTRY SURFACE, 64 commits, 35 percent rewritten. Settled once the endpoint contract stabilised.
  • mcp/tools/search_tools.py — RETRIEVAL BINDINGS, 58 commits, 56 percent rewritten. The newest subsystem, and the most active relative to its age.
  • services/github_service.py — REMOTE ACQUISITION, 39 commits, 18 percent rewritten. Effectively finished — a boundary that stopped moving.

ARIA understands not only what the repository is, but how it changes.

MEMORY → REASONINGStructure and history become context.

06 — GROUNDED RETRIEVAL

Ask the codebase.

Conversational responses grounded in the repository's symbol graph, dependencies, and source files.

STANDBY
GROUNDED REASONING ENGINE
QUERY

GRAPH TOPOLOGY CONTEXT
Topology depth
12
Graph directed
DiGraph
Nodes evaluated
1,284
Cycles found
0

ARIA does not retrieve text that resembles the question. It walks the symbol graph, collects the modules that actually participate, and answers from those.

ARIA answers from repository structure, not just text similarity.

REASONING → READING PATHContext tells you where to begin.

07 — ONBOARDING

Context tells you where to begin. Centrality-ranked file sequences help engineers understand the architecture before diving into implementation details.

ARCHITECTURAL ONBOARDING

Start where the architecture says the code matters most — not at the top of the alphabet.

01/ 05
~40 MIN TOTAL
ROOT ORCHESTRATION HUB

Start here: every analysis stage is sequenced from this module.

PAGERANK 0.94·~12 MIN READ

ILLUSTRATIVE RANKING

  1. 01

    services/analysis_orchestrator.py

    ROOT ORCHESTRATION HUB·~12 MIN
    0.94
  2. 02

    core/ast_parser.py

    SYMBOL EXTRACTION CORE·~9 MIN
    0.82
  3. 03

    core/dependency_graph.py

    TOPOLOGICAL ENGINE·~8 MIN
    0.76
  4. 04

    models/schemas.py

    CONTRACT LAYER·~5 MIN
    0.72
  5. 05

    backend/routers/repositories.py

    HTTP ENTRY SURFACE·~6 MIN
    0.68

READING PATH → PIPELINEUnderstanding starts with deterministic structure.

08 — PIPELINE ARCHITECTURE

Each stage executes as an isolated analyzer. The graph is computed, indexed, and exposed as structured engineering intelligence.

STAGE 01 · CLONE
  1. 01

    CLONE

    Target repository

    Shallow clone into an ephemeral workspace. Nothing is written back.

  2. 02

    DETECT

    Ecosystem & stacks

    Languages, build manifests, package boundaries and entry points.

  3. 03

    INDEX

    Symbol extraction

    Tree-sitter parses each file into fine-grained symbol hierarchies.

  4. 04

    ANALYZE

    Graph computation

    Centrality, cycle detection and blast radius over the NetworkX graph.

  5. 05

    ANSWER

    Knowledge retrieval

    Graph-grounded retrieval that cites the files it reasoned over.

PIPELINE → INVESTIGATION Simulated change traced live through ARIA's dependency graph.

09 — FORENSIC INVESTIGATION

Simulated change. Deterministic proof.

Inspect how ARIA resolves syntax tokens, traverses callers, isolates blast radius, and issues review actions before the merge.

SIMULATED INVESTIGATION PIPELINE
PRESETS:
STAGE 1/5 · CHANGE SCENARIOSTORAGE · DRIVER

Fix SQLite timeout handling

Add exponential backoff and lock retry to SQLite store connector during concurrent ingest.

SYNTAX DIFF CLUSTERAST NODES (LEXICAL PARSE)
- timeout = DEFAULT_TIMEOUT
+ timeout = calculate_exponential_backoff(attempt, max_retries)
+ retry_on_exception(SQLiteTimeoutError)

ARIA detects modified syntax tokens directly from the AST diff without relying solely on raw text patches.

EVIDENCE SYNTHESISTOPOLOGY ENGINE
CHANGE ORIGIN
Δ ast_diff
TARGET SYMBOLS
3 resolved
AFFECTED REACH
2 files
VERIFIED DECISION
XS · 69%
Open Live Studio Lens→
"ARIA traces the relationship graph before the merge — not inferred from raw diff text."
09 — TECHNOLOGY
PLATFORM CAPABILITIES
STRUCTURE
  • › AST parsing
  • › Repository topology
  • › Dependency graphs
  • › Cycle & layer detection
SEMANTICS
  • › Symbol index
  • › Call graph resolution
  • › Centrality ranking
  • › Deterministic AST scopes
CHANGE
  • › Change impact analysis
  • › Dead code detection
  • › PR blast radius & risk
  • › Architectural drift
MEMORY
  • › Git history telemetry
  • › Commit concentration
  • › Code churn tracking
  • › Hotspot identification
INTERACTION
  • › ARIA Chat (grounded RAG)
  • › Guided Reading Path
  • › Model Context Protocol (MCP)
  • › REST API & CLI

CORE ENGINE

05
  • 01 Tree-sitter AST 01/05
  • 02 NetworkX Topology 02/05
  • 03 FastAPI & SSE 03/05
  • 04 Python 3.12 04/05
  • 05 SQLite & ChromaDB 05/05

INTELLIGENCE

04
  • 01 Gemini 2.5 Flash 01/04
  • 02 DeepSeek fallback 02/04
  • 03 Deterministic RAG 03/04
  • 04 Incremental Index Caching 04/04

INTERFACE & CLIENTS

04
  • 01 Astro + React Islands 01/04
  • 02 TypeScript 02/04
  • 03 Tailwind CSS 03/04
  • 04 MCP Server protocol 04/04

No inference where a parser will do. Language detection, symbol extraction and graph computation are deterministic; models are used only to describe what the graph already proves.

Stop reading the codebase.

Let ARIA map it first.

OPEN ARIA → VIEW SOURCE ↗ SELF-HOSTED · OPEN SOURCE · MIT LICENSE
10 — ENTER THE SYSTEM

Start with a repository.

Paste a public GitHub repository and let ARIA map its structure, relationships, and engineering intelligence.

SYSTEM COMMAND
DETERMINISTIC AST PARSER · GRAPH INTELLIGENCE · ZERO WRITEBACK
SAMPLE CODEBASESClick to load