Three surfaces, one evidence file

Start local. Stay local as long as you want.

Every surface below reads the exact same air.json. Nothing is re-derived, re-guessed, or inconsistent between them.

$ aletheore scan

Build the evidence

Tree-sitter and git history produce air.json and air.toon. No LLM in the scan path and no account. A scan makes dependency advisory and license lookups and checks for a newer release, each of which can be turned off, and sends a crash report only if the CLI itself crashes.

aletheore scan .

$aletheore scan .
Scanning /path/to/repo...
  → Detecting languages, frameworks, and build tools
  → Building module dependency graph (parsing source with tree-sitter)
  → Analyzing git history and ownership
  → Scanning working tree for secrets
  → Scanning git history for secrets (can be slow on large histories)
  → Clustering modules and checking layer conventions
  → Mapping API endpoints
  → Detecting dead code
  → Computing git hotspots
  → Checking dependencies for known vulnerabilities (OSV.dev)
  → Checking dependency licenses (one registry lookup per pinned dependency)
  → Running static analysis scanners
  → Checking dependency licenses: 1/196 ... 196/196
✓ Scan complete
  Evidence written to .aletheore/air.json

on: pull_request

The GitHub App keeps watch

Automatic PR comments, branch-protection checks, and Flash reviews on every push: same evidence, hosted.

on: pull_request

$push to an open pull request
  review    Flash review of the changed files, grounded in the evidence
  secrets   a new secret fails the branch-protection Check Run
  alerts    Slack / Teams message on new findings (AIR)
  status    live endpoint checks, each tied to its source handler

$ aletheore mcp .

Agents query it directly

Read-only tools over stdio, TOON-encoded: a coding agent gets exact answers instead of re-reading files on every lookup.

aletheore mcp .

$aletheore mcp .
aletheore_overview                      Navigate
aletheore_neighborhood                  Navigate
aletheore_get_blast_radius              Navigate
aletheore_find_evidence_for_endpoint    Evidence
aletheore_ownership                     History
aletheore_hotspots                      History
aletheore_search_codebase               Search
aletheore_symbol_source                 Search
aletheore_verify_citations              Evidence
... 35 more, all read-only

aletheore scan .

$aletheore scan .
Scanning /path/to/repo...
  → Detecting languages, frameworks, and build tools
  → Building module dependency graph (parsing source with tree-sitter)
  → Analyzing git history and ownership
  → Scanning working tree for secrets
  → Scanning git history for secrets (can be slow on large histories)
  → Clustering modules and checking layer conventions
  → Mapping API endpoints
  → Detecting dead code
  → Computing git hotspots
  → Checking dependencies for known vulnerabilities (OSV.dev)
  → Checking dependency licenses (one registry lookup per pinned dependency)
  → Running static analysis scanners
  → Checking dependency licenses: 1/196 ... 196/196
✓ Scan complete
  Evidence written to .aletheore/air.json

CLI workflow

Scan once. Query the evidence many ways.

The deterministic scan writes AIR, Aletheore's Intermediate Representation. Downstream commands read AIR instead of scanning blindly again.

$pip install Aletheore
$aletheore scan .
$aletheore query endpoints
$aletheore dashboard

Real output from scanning the Aletheore repository itself. The path is shortened and the one-per-package licence progress lines are collapsed to one; nothing else is edited. The query kinds are the CLI's own listing for the released version.

Install and scan

pip install Aletheore
aletheore scan .

Builds .aletheore/air.json, .aletheore/air.toon, and history snapshots.

Audit locally or managed

aletheore audit .
aletheore audit . --managed

Use your own provider locally, or run a managed audit with an Aletheore API token.

Search and dashboard

aletheore index .
aletheore query search-codebase "auth flow"
aletheore dashboard .

Build a semantic index (on your machine by default), ask targeted questions, and inspect evidence visually.

MCP tools

An agent gets exact answers, not a re-read of your repo.

Aletheore's MCP server exposes repository evidence as structured, TOON-encoded tool results. Agents can ask for imports, owners, endpoints, symbols, semantic search, source evidence, and managed audits without pasting the whole repository into context.

Navigate 5

Structure 6

Security 4

Runtime 5

History 4

Evidence 4

Search 5

Hallbook 7

Run 4

AI-assisted 1

Other 1

Navigatedefault

aletheore_overview

A repo-level summary: languages, frameworks, monorepo structure, dependency-graph size, module/cluster counts, and git age/commit cadence/branch count.

44 tools are registered by default in aletheore 0.12.0. aletheore_answer is added when the server is started with --agent, and aletheore_managed_audit only with explicit consent to transmit evidence externally (ALETHEORE_MCP_ALLOW=...,external).

Evidence contract

One source of truth.

Everything downstream reads the same deterministic evidence. AIRview, the dashboard, query, MCP, and audits do not invent files, routes, symbols, or owners that are not present in AIR.

.aletheore/

air.json

Canonical repository evidence: modules, symbols, dependencies, endpoints, architecture, security, ownership, and history.

air.toon

TOON-encoded evidence for agent-facing and model-facing contexts where token cost matters.

history/

Snapshots used for changes, drift, and continuity between scans.

Everything below reads it

query
dashboard
mcp
audit
AIRview
PR reviews

Depth without mystery

Technical surfaces that stay evidence-first.

AIRview
Architecture maps
AI-written subsystem pages and dependency diagrams grounded in deterministic architecture clusters.
MCP
Agent context savings
Agents request focused evidence instead of dragging entire repositories into model context.
Security
Risk resolution
Secrets, vulnerabilities, dependencies, endpoints, and infrastructure findings trace back to source.
Health
Endpoint checks
aletheore healthcheck runs GET-only checks against mapped API endpoints and saves the result locally.
Local
Privacy by default
The CLI scan runs locally. Managed and hosted flows use derived evidence according to the selected workflow.