01 / ecosystem · code intelligence

your codebase,
mapped.

atlas scans your Python project and builds a call graph, dependency tree, and risk map — deterministically. No LLM, no API calls, no telemetry.

28s · 221 files · 1563 symbols · offline forever

get atlaspip install atlas-intel
100% offline

no API calls, no telemetry

LLM-free

same input → same output, always

pip installable

windows · linux · macos

HTML report

one self-contained .html file

call graph · risk overlay · 8 nodes shown
risk ≥ 70 risk ≥ 40 ok
api.routes61auth.validate_token43order.create_order87payment.charge_carddb.get_usercache.lookup82stripe.chargediscount.apply
71.4/100
health
2
critical hotspots
3
security issues

live from atlas scan . · nodes colored by risk_score · edges from call graph

0
API calls
28s
221-file scan
12+
analysis modules
100%
offline
// the report
atlas_report.html — one file, no internet
system_map · call_graph.svg — dependency topology
api.routes→ auth.middleware, order.handler, payment.service
auth.middleware→ db.users, cache.sessions
order.handler→ payment.service, inventory.check
payment.servicerisk: 43→ stripe.client
stripe.client⚠ risk: 87 · 3 callers
1563 nodes · 4712 edges · 2 circular dependency chains detected
// critical
stripe.charge
risk: 87 · hardcoded secret path
payment.charge_card
risk: 82 · no rate limit · 41 callers
// high
auth.validate_token
risk: 61 · cyclomatic: 14
order.create_order
risk: 43 · 6 downstream deps
// health
71.4/100
Fair — 3 modules need attention
complexity
68%
duplication
23%
dead code
11%
// atlas ask
natural language · deterministic · 0 tokens

query your codebase
without spending a token.

atlas ask runs a deterministic NLP engine against your code graph. Intent detection, entity extraction, BFS traversal — no LLM, no API key, instant.

atlas ask "describe the project"→ overview: architecture summary + top-level modules
atlas ask "why is validate_token risky?"→ risk: complexity score + upstream callers + security flags
atlas ask "what breaks if PaymentService changes?"→ impact: BFS downstream — N connected symbols
atlas ask "dead code in auth module"→ orphan: unreachable functions with no callers
atlas ask "why is validate_token risky?"
$ atlas ask "why is validate_token risky?"
ATLAS Intelligence — RISK
auth.validate_token (auth/middleware.py:47)
Risk score: 61/100
Cyclomatic complexity: 14 (threshold: 10)
Centrality: 0.847 — high blast radius
Callers (14 upstream):
· api.routes.get_user (routes.py:23)
· api.routes.create_order (routes.py:58)
· ... 12 more
Security flags:
· hardcoded JWT_SECRET fallback detected
· bare except clause swallows token errors
█
// modules
see tiers →
Core Engine
FREE

AST parser (libcst), graph builder (rustworkx), HTML reporter. Included in all tiers.

System Map
SOLO

Dependency graph. Import chains, circular deps, module clusters.

Risk Radar
SOLO

Hotspot detection. Cyclomatic complexity, churn rate, change coupling.

Security Shield
SOLO

Static security analysis. SQLi, XSS, secret exposure, unsafe patterns.

Code Health
SOLO

Maintainability index, dead code detection, duplication ratio.

Signal Map
PRO

Cross-module signal propagation. Which changes affect which components.

Decision Center
PRO

Rule-based architectural decisions. Policy enforcement, guardrails.

Ownership Map
PRO

Bus factor analysis, knowledge silos, git blame aggregation.

Rewind
ENT

Snapshot diffing. Compare any two scan results, track regression.

What-If
ENT

Impact simulation. Predict risk before you refactor or delete.

Commit Guard
ENT

Pre-commit gate. Block commits that violate health or security thresholds.

Architecture Intelligence
ENT

Layer violations, coupling analysis, architectural anti-patterns.

// atlas mcp
stdio + HTTP · 11 tools · Claude Code · Cursor · CI

every module
machine-consumable.

atlas mcp --project . starts a stdio MCP server. Every analysis module becomes a callable tool for Claude Code, Cursor, or any MCP host. Nothing is UI-only.

Claude Code / Cursor (stdio)
atlas mcp --project .
HTTP (multi-device, CI)
atlas mcp --project . --transport http --pre-scan
claude_desktop_config.json
{"command":"atlas","args":["mcp","--project","."]}
11 MCP tools exposed · atlas mcp --project .
scan_projectFull scan + graph build on demand
top_risksHighest risk_score nodes in the graph
detect_violationsSecurity issues, arch violations, dead code
simulate_changeBlast radius: what breaks if X changes?
pack_contextAI-ready context pack for a query/issue
trace_callsCall chain from root to leaf
dead_codeUnreachable functions and orphan classes
security_issuesBandit + custom pattern matches
graph_summaryNode count, edge count, health score
node_infoFull signals for a single symbol
module_outputRaw output of any loaded module
atlas surgeon → deepstrain

atlas surgeon validate_token generates a surgical refactor prompt from the live call graph — callers, callees, risk scores, existing patterns. Send that prompt to deepstrain eval and it executes the change with full tool access.

atlas scan → graph → atlas surgeon → surgical prompt → deepstrain eval → code change
// quick start
01pip install atlas-intel
02atlas activate-file ~/Downloads/atlas_license.json
03atlas scan .
also: atlas ask "describe the project" · atlas context "auth bug" · atlas mcp --project .
// pricing

Solo

$19/mo

indie dev

core + system_map + risk_radar
security_shield + code_health
atlas ask + atlas context
MCP server (stdio + HTTP)
get solo

Pro

$39/mo

team of 5

everything in Solo
signal_map + decision_center
ownership_map + what_if
atlas export (portable ZIP)
get pro

Enterprise

$99/mo

org-wide

everything in Pro
architecture_intelligence
commit_guard + rewind
unspaghetti (refactor assistant)
get enterprise

// ecosystem

atlas maps · deepstrain ships · adauto narrates

atlas tells you what your codebase looks like — deterministically, offline, with no tokens. deepstrain changes it. adauto turns the change into distribution.

bundle — save 20% all products