Axiom Inspector

[ AxiomCore · Documentation ]

Axiom Inspector

Query what an Axiom application declares, resolves, can access, executed and changed, from the CLI or a local dashboard.

Axiom Inspector turns compiler, resolver, package, authority and runtime evidence into a queryable model of your application. Use it to answer four questions:

  1. What is this application?
  2. What can it do, and why is it allowed to?
  3. What happened during one captured execution?
  4. What would a change affect?

It works on frontend-only, backend-only or combined workspaces.

Evidence becomes an explorable model.

Begin with available evidence.

Acore source, verified contracts, locks, packages, authority, and captured runtime events contribute different evidence. Missing evidence remains explicit.

Evidence sources
Source spansArtifact identities
Conceptual model · This illustration is not a live runtime trace.

The CLI, machine formats and dashboard all query the same evidence engine with the same semantic IDs and graph revision. The dashboard does not reimplement the compiler, and an AI explanation never becomes a new source of facts.

Axiom Inspector dashboard Overview for the task app: 100 semantic facts, 155 relationships, three targets, no extensions, no readiness blockers, and an architecture graph from task-app to its frontend modules, Home page and operations

The dashboard Overview for the documentation task app, served by CLI 0.149.1.

Start locally

From an application workspace:

Terminal
axiom inspect overview .
axiom inspect serve .

The first command prints a summary of every discovered fact. The second starts the loopback dashboard. For agents and CI, add --format json (one complete result) or --format jsonl (one record per line).

What you can inspect

AreaCommandsExamples
Workspaceoverview, source, serveEvery discovered fact, one saved source file as the editor sees it, the dashboard
Frontendfrontend, ui, state, styles, accessibilityRoutes, pages, components, primitives, actions, state, queries, mutations, styles, responsive branches, assets, accessibility
Backendbackend, operation, data, exposure, cache, serverServices, operations, endpoints, models, fields, projections, validation, auth, errors, cache, retry, streams, implementation bindings, private server plans
Contracts and packagesdependencies, contract, executable-closure, provenanceDirect and transitive imports, immutable identities, target support, provenance, executable closure; extension exports, digests, permissions, limits and source availability
Securitysecurity, authority, finding, readiness, accessExposure, requested, granted and effective authority, findings, blockers
Relationshipsnode, evidence, trace, why, reachability, lineage, graph, query, target-compareTyped paths between facts, per-target differences
Runtimerecord, runtime, export-audit, import-auditRedacted sessions, traces, extension lifecycle, denials, state changes, cancellation, teardown, retention
Changessnapshot, validate-snapshot, diff, impactSemantic changes, approval classes, bounded impact paths
Questions and handoffask, answer, export-inspector, import-inspector, release-checkBounded questions, signed offline exchange, Inspector self-checks

The axiom inspect reference lists all 46 subcommands.

Inspector reports evidence, not guesses

Inspector states only what its current evidence establishes. It does not infer an external backend implementation from an endpoint name, label static reachability as an observed call, or treat requested extension authority as a grant. When evidence is unavailable or a traversal is truncated, that limitation is part of the result.

Inspector keeps its local state under .axiom/inspector/ in the workspace: a query cache, snapshot copies, runtime sessions, planner receipts and the live connection descriptor. Inspection makes no hosted upload.

  1. Run axiom inspect overview . after adding or importing a contract.
  2. Review dependencies, executable closure, security and authority before accepting third-party content.
  3. Trace important actions and operations in both directions, with the dashboard open while you develop.
  4. Capture runtime evidence only when a question needs observed behavior, and gate CI on semantic diff between snapshots.

Next steps