AxiomGate

The description states that AxiomGate is a "local-first governance runtime for OpenAI Codex." It operates through five governed stages:

1. Plan: Objective and acceptance criteria compile into a hashed contract before Codex writes a line.
2. Guard: Identity resolved, policy enforced at the Codex hook, not suggested in a prompt.
3. Run: Codex builds under a sandbox and an intent boundary, with usage ledgered.
4. Verify: Tests and scanners produce machine evidence; model claims are inadmissible.
5. Prove: Completion is gated on evidence and sealed in a tamper-evident Build Receipt.

AxiomGate generates receipts that record the contract hash, a SHA-256 chained evidence trail, criterion citations, and the completion verdict. These receipts can be verified offline with no account required.

The system integrates through native Codex surfaces rather than wrapping them:

• Hooks for PreToolUse and PermissionRequest enforcement
• An MCP server exposing six tools (read-only except approve)
• A Codex plugin and marketplace manifest
• A custom read-only verifier agent

It also uses PatchPilot for dependency verification, and approvals reach a phone over Telegram using Bot API long polling only.

Inference: The product appears to be a framework or toolset that enforces governance rules on Codex-based agents by introducing checkpoints and evidence-based validation at each stage of execution.

The description claims AxiomGate addresses a gap in current agent behavior: "nothing checks whether they stayed inside the boundary you intended, or whether 'done' was ever true."

It positions itself as a solution to:

• Action-boundary violations (e.g., 55.8–67.8% failure rate reported by UnderSpecBench)
• Misalignment between model claims and actual execution
• Lack of enforcement when permissions live in editable config files

The evolution from claim to product is framed around:

1. Identifying the root problem: "Permissions usually live in a config file the agent itself can edit"
2. Solving it with a runtime that enforces policy at the hook level
3. Ensuring evidence-based validation ("Model claims are inadmissible")
4. Providing tamper-evident receipts

Inference: The positioning has shifted from a general critique of AI agents to a specific technical architecture aimed at enforcing boundaries and validating outcomes.

The description does not name specific customers or use cases beyond the context of OpenAI Codex. It implies that users are developers or teams who:

• Use OpenAI Codex for automation
• Want to enforce strict governance over agent actions
• Require verifiable completion proofs

It suggests a focus on enterprise or advanced developer workflows where compliance and auditability matter.

Inference: The target ICP likely includes developers or engineering teams working with AI agents in regulated environments or those requiring high assurance of execution integrity.

There is no evidence of pricing, revenue models, or monetization strategies in the project description. The author states that AxiomGate is built using Codex end-to-end and integrates with native surfaces, but does not describe how it would be sold or consumed commercially.

Not evidenced: No indication of business model, pricing tiers, or customer acquisition strategy.

The system uses:

• Native Codex hooks (PreToolUse, PermissionRequest)
• MCP server exposing six tools
• Codex plugin and marketplace manifest
• A custom read-only verifier agent
• SHA-256 chained evidence trails
• Dependency verification via PatchPilot
• Telegram Bot API for approvals

It supports deterministic regressions:

• Wrong-target ownership denial
• Exact-command approval binding
• Missing-evidence completion blocking

The system is designed to be installable and usable without login or credentials after initial setup.

Inference: The technical approach involves deep integration with Codex internals, sandboxing, and cryptographic chaining for evidence integrity.

There is no evidence of traction, customers, or adoption beyond the authors’ own demonstrations. The project was submitted to an OpenAI hackathon, indicating early-stage development.

The description mentions:

• A 23-case regression suite
• Fixing a critical bypass in their own system
• Demo orchestrator labeling scenes as REPLAY, SAMPLE, or STORED LIVE EVIDENCE

Not evidenced: No data on usage, revenue, customer base, or product maturity beyond prototype-level functionality.

The description does not mention competitors or direct market comparisons. It focuses on the problem of agent overreach and lack of enforcement rather than positioning itself against existing tools.

Inference: The space is likely emerging, with no clear category leaders yet. AxiomGate may be addressing a niche in AI governance for automated agents.

1. No real-world usage or adoption: The project is described as a hackathon submission with no evidence of production use.
2. Self-reported claims without corroboration: Many assertions (e.g., violation rates, enforcement effectiveness) are unverified.
3. Limited team size (2 members): Suggests limited capacity for scaling or commercial development.
4. No pricing or monetization strategy: Indicates no clear path to revenue generation.
5. Dependency on Codex only: If Codex becomes obsolete or changes APIs, AxiomGate may not be portable.

Inference: The risk of misalignment between stated goals and actual product viability is high due to lack of external validation or traction.

1. What specific real-world use cases have you tested AxiomGate in?
2. How does it handle edge cases where Codex behavior deviates from expected patterns?
3. Have you validated its effectiveness against actual agent deployments, not just demos?
4. Is there a plan to support other AI agents beyond Codex?
5. What is the roadmap for commercialization or monetization?
6. How do you intend to scale beyond the current two-person team?