i2x
See it work ↘

Reliability for Codex & Claude projects

Your agent writes the change.
Know what happens next.

Keep your project and model account where they are. Ask Codex or Claude normally; i2x gives the agent bounded work, checks the returned result, and shows you what is proven, uncertain, or waiting for approval.

● In development · local product preview

Illustrative product view · request portalexact change · 7e91…c8a2
Current workflowVerify checkout update
Checking
✓✓345
UnderstandProtectRehearseDecideVerify
Protected promiseA signed-in user can submit a request
4 / 5 checks
✓

BuildExpected 0 · actual 0

verified
✓

AuthorizationWrong user denied

verified

PersistenceReading stored result

running
Source7e91…c8a2
Origini2x runner
Evidencesigned
Your agent and subscriptionExact-change evidenceHuman approval before external effects

Between “done” and dependable

Coding got faster.
Confidence did not.

Codex or Claude can turn an idea into working software. Then the difficult questions arrive.

  • 01What did this change affect?
  • 02Did the important behavior still work?
  • 03Is this an agent report—or independent evidence?
  • 04If it fails, what is the safe next move?

i2x makes those questions one visible, repeatable workflow.

One project. One continuous story.

From agent change
to retained proof.

Choose a stage to see what you see, what the system commits, and how failure returns safely to the workflow.

01 / 06

See the project as a system.

i2x maps supported code, dependencies, declared flows and limitations before anything is changed.

You seeA readable project map and what remains unknown.
The system commitsAn immutable inventory and governed understanding record.
projectrequest-portalsource · a912…4f10
interfaceweb appdeclared
statedatabasedetected
identitysessionsneeds promise
FAIL → HOLD → REPAIR → RERUN

No blurred meanings

“Finished” is not the same as “verified.”

i2x keeps three truths separate so a green badge cannot promise more than the evidence supports.

A
Agent report
Analysis complete

✓Code review completed

!Limitations retained

Not independently verified

Agent report

Useful finding

Your agent completed a bounded task and returned a schema-checked report. It can guide work, but it does not verify itself.

Task completed · not independent
B
Evidence receipt
Source-bound evidence

ExpectedPASS

ObservedPASS

Digest and provenance inspectable

Evidence receipt

Exact observation

The result binds source, environment, expected and actual values, checks, exclusions and provenance under one digest.

Evidence authentic · inspectable
C
Current assurance
Current passing evidence

✓Exact revision

✓Certified runner

Valid until source or evidence changes

Current assurance

Verified now

A certified runner produced passing, still-current evidence for the exact revision. Expiry or changed source removes the claim.

Current passing assurance

Bring your subscription

Keep coding where you already code.

Use your own Codex or Claude subscription through its official client. A revocable workspace connector gives the agent bounded i2x tasks; i2x does not resell model access or take over your provider account.

i

An honest boundary: i2x can prove the connector round trip works. It cannot see your provider plan or remaining allowance.

config.toml
[mcp_servers.i2x]
command = "node"
args = ["i2x-mcp"]

[mcp_servers.i2x.env]
I2X_CONNECTOR_TOKEN = "••••••••"
MCP round trip provedWorkspace-scoped · revocable

Choose how you work

Three ways in.
One clear result.

You keep using the tools you know. i2x gives each request a project boundary, tells the agent exactly what it may do, and returns a result that separates the agent’s report from independently checked evidence.

01 · ON YOUR COMPUTER

Use Codex or Claude locally

Best when your project is a folder on Windows, macOS, or Linux.

  1. Open the project in your usual terminal or editor.
  2. Start Codex CLI or Claude Code with your own subscription.
  3. Connect the i2x MCP tool and choose read-only or read-write access.
  4. Ask normally. Your agent receives bounded i2x tasks and returns structured results.
You receive

A project map, protected behaviors, check results, evidence level, and the next safe action.

02 · FROM A REPOSITORY OR CI

Run checks without a desktop session

Best for GitHub, GitLab, Bitbucket, a private Git server, or your CI pipeline.

  1. Connect only the repository and branch or commit you choose.
  2. Select your runner, CI worker, or an isolated i2x runner.
  3. Your pipeline calls the authenticated i2x API for a bounded job.
  4. i2x returns machine-readable status and a human-readable evidence receipt.
You receive

The exact source checked, pass and fail details, unresolved limits, and a result your workflow can verify.

03 · FROM YOUR PRODUCT

Call i2x through the API

Best when your web app, CLI, automation, or agent already owns the user experience.

  1. Create a revocable workspace connection and project scope.
  2. Send the goal and exact source reference through HTTPS.
  3. Your agent performs the requested reasoning; i2x enforces the allowed workflow and required outputs.
  4. Poll or receive the governed result, then show it in your own interface.
You receive

A stable response with lineage, evidence level, limitations, and any approval still required.

You control

Project access, credentials, network permission, source version, and final approval.

Codex or Claude does

Project-specific reasoning and coding through the subscription or account you chose.

i2x does

Scopes the work, applies the required checks, records lineage, and labels what is—and is not—verified.

Important: connecting i2x does not give an agent permission to approve its own result, publish software, or start recovery. Those effects remain separate and require your approval.

You choose the boundary

Your project.
Your way of working.

Start with one simple decision: where should your code stay? i2x adapts around that choice. You do not need to move your repository, buy a VPS, or give i2x more access than the work requires.

01

Local folder

Code and credentials stay on your computer. Codex or Claude works through the local runner.

Your computer → local runner → governed result
02

GitHub / GitLab repository

Select a repository and choose read-only analysis or a read-write worktree.

Git repository → chosen runner → governed result
03

Your server or cloud

Run i2x in your VM, container, development environment, or CI pipeline.

Your infrastructure → your policy → governed result
04

Temporary i2x clone

Grant revocable access for isolated execution without buying or managing infrastructure.

Scoped access → isolated runner → retained evidence
05

Offline evidence package

Keep the source offline and export a content-bound package for verification.

Offline project → evidence package → portable verification

Windows · macOS · Linux · CI and remote runners · web control from iOS and iPadOS

Compare every setting

Available foundationPlannedLater

Project source

Local folder Now
GitHub / GitLab / Bitbucket Planned
Self-hosted Git repository Planned
CI workspace Planned
Your server or cloud Planned
Customer-hosted control plane Planned
Temporary i2x clone Planned
Offline evidence package Planned

Execution location

Local runner Now
Container or devcontainer Planned
Your CI pipeline Planned
Your server or cloud runner Planned
Customer-hosted i2x runner Planned
i2x hosted runner Planned

User interface

Terminal / CLI: PowerShell, Command Prompt, bash, zsh Now
VS Code and compatible editors Planned
Cursor and Windsurf Planned
JetBrains IDEs Planned
i2x web portal Planned
CI automation Planned
Mobile web control Planned

Operating system

Windows local runner Now
macOS local runner Planned
Linux local runner Planned
iOS / iPadOS web control Planned
Any OS through CI or remote runner Planned

AI access

Your Codex or Claude subscription Now
Your provider API account Planned
i2x-managed enterprise account Later

Repository permission

Read-only analysis Choice
Read-write worktree Choice

Credential custody

Credentials stay on your device Now
Your vault integration Planned
i2x credential vault Later

Network policy

No network access Choice
AI provider only Choice
Restricted allowlist Planned

What every customer gets: connect the project in the way that fits your environment, let Codex or Claude build a readable system map, identify behavior that must not break, run bounded checks, and see clearly whether each result is an agent report or independent evidence. You choose the project boundary, credentials, network access, and read-only or read-write permission. Access is project-scoped and revocable; release and recovery always require separate human approval.

The operating model

01

Your project stays the subject.

Every plan, task and receipt is bound to the project, tenant and exact source—not a generic AI conversation.

02

Agents propose. Kernels decide.

Deterministic controls enforce coverage, evidence, authority and state transitions. Agent confidence is never the policy.

03

Effects wait for a human.

Release and recovery require current evidence, exact scope and fresh human approval before any supported external effect.

04

Experience improves skills safely.

Redacted, recurrent outcomes may propose a revision. Evaluation, consented canary and rollback come before promotion.

Pricing

Compare every plan.
Start free.

See exactly what each plan includes. Every plan keeps model access separate: use your own Codex or Claude subscription.

CapabilityFree$0LocalPersonal$19 / monthUp to 10 projectsShared$49 / monthUp to 500 projects
Project access
Local folders on Windows, macOS and Linux✓Included✓Included✓Included
Codex or Claude subscription connector✓Included✓Included✓Included
Read-only or read-write project access✓Your choice✓Your choice✓Your choice
Git repository connection—Not included✓Included✓Included
CI, server and remote-runner modes—Not included✓Included✓Included
Offline evidence package✓Included✓Included✓Included
Understanding and protection
Readable project system map✓Included✓Included✓Included
Protected behavior plan✓Included✓Included✓Included
Governed specialist-agent workflow✓Included✓Included✓Included
Failure handoff and bounded repair loop✓Included✓Included✓Included
Evidence and assurance
Locally reported results✓Included✓Included✓Included
Portable evidence export✓Included✓Included✓Included
Independent hosted verification—Not included✓120 min / month✓600 min / month
Retained evidence storage—Local only✓1 GiB✓5 GiB
Release and recovery evidence gates—Not included✓Included✓Included
Workspace control
Project capacity✓Local projects✓Up to 10✓Up to 500
Personal activity history✓Local✓Included✓Included
Workspace members and roles—Not included—Not included✓Included
Shared approval and audit trail—Not included—Not included✓Included

i2x · intelligence to execution

Build with agents.
Move with evidence.

A reliability layer for newcomers who want to keep improving what they built—without pretending every agent answer is proof.

Walk through the product ↑