Skip to main content

Developed by Newport Resonance

Know when your AI session is still worth trusting.

ToM Bar is a live session-assurance monitor for Codex and Claude Code. It checks whether the accepted objective still governs the work, identifies what remains failed, stale or unproven, and shows the next issue to inspect before a weak result becomes a bad decision.

Form
Native macOS companion
Sessions
Codex and Claude Code
Control
User-reviewed actions
ToM Bar Sessions screen prioritising concurrent Codex and Claude Code sessions by the issue that needs attention.
Deterministic product prototype; illustrative session and usage data.

The assurance gap

AI activity is not the same as reliable progress.

Agents can work for hours, compact context, launch sub-agents and produce a convincing completion claim. The harder question is whether the current session still reflects what the user asked for—and whether the result has current proof.

Attention

See what needs inspection.

Prioritise concurrent sessions and name the current issue: a failed gate, missing proof, stale evidence, possible drift or unsafe continuity.

Basis

Understand the observed evidence.

Keep judgments tied to what ToM Bar could actually observe, with partial states when analysis coverage or provenance is insufficient.

Action

Recover without losing the objective.

Prepare a bounded follow-up or completion plan, then wait for fresh evidence before reporting that the issue improved.

Product walkthrough

Follow the line from accepted intent to bounded recovery.

Use the four steps to inspect how ToM Bar keeps context, assurance, proof and usage distinct. The controls navigate this explanation; they do not operate or simulate the product.

ToM Bar Context screen showing an accepted objective, completion criteria, constraints and session-specific context anchors.

Walkthrough 01

Keep the objective connected to the work.

Inspect the objective, criteria, constraints and decisions that currently govern one session. Context remains session-specific, even when several agents are active.

Continuity measures describe whether accepted intent survives in the observed work. They do not certify that the task is complete or correct.

Deterministic product prototype; illustrative session and usage data.

Evidence before confidence

Three signals. Three different questions.

ToM Bar is designed to help people decide whether to rely on an observed session. It does not compress assurance, completion and consumption into one impressive-looking score.

Session Assurance

A ToM-owned view of observed continuity, evidence reliability, goal alignment and recovery stability. It is not a probability that an answer is correct.

Completion proof

Tests, builds, sources and criterion-bound evidence remain visible as their own question. Healthy continuity does not prove the work is complete.

Provider usage

Quota and context-window consumption are shown only when the provider exposes them. Usage does not raise or lower Session Assurance.

Private by design

Bounded observation. User-controlled action.

ToM Bar is designed for local, privacy-conscious operation. Durable continuity uses encrypted derived metadata, hashes and evidence references rather than a stored copy of the conversation. Provider authentication remains with the provider integration.

  • No universal correctness claim.
  • No silent provider prompt submission.
  • No unobserved result presented as verified.

Explore the product direction

Bring a stronger basis for trust to serious AI-assisted work.

Talk with Newport Resonance about the ToM Bar product direction, live-session assurance and the broader ToM cognitive intelligence system.