AgentPort 3.1.30

AgentPort 3.1.30

Haiku-only usage protection now blocks serialised requests larger than 256 KiB before forwarding, including repeated retries and both Codex compaction paths. The error identifies instruction, conversation and tool sizes. This safeguard does not truncate context or change native GPT settings. It is a protective byte limit, not a token or billing estimate, and may block legitimate long Haiku conversations. The underlying context inflation and ordinary-session cost efficiency remain under investigation. No paid provider request was made during verification.

Routed names in Codex lead with the provider, for example (OpenCode Go) Claude Haiku 5.5 and (OpenRouter) Qwen. Providers and models sort alphabetically in API Models, with confirmed free models grouped first. Certified Codex subagent routes retain priority slots so agent spawning remains available. Existing visibility selections are preserved.

NVIDIA NIM setup is available at API Models > Connect provider > NVIDIA NIM. Enter the NVIDIA key there, refresh the catalogue and select models. The provider listing does not imply free quota or access to every model.

Verified: 165 routing/catalogue/discovery regression tests; 107 desktop tests, with six platform/environment skips; syntax and TypeScript checks. Local mock integration tests confirmed zero upstream calls for oversized requests, retries and compaction, and one upstream call for a small request.

Available for Windows x64 and Apple Silicon macOS. The Mac preview supports Codex and API providers; local inference is not included. It is ad hoc signed, not Apple-notarised. Fully quit and reopen Codex after updating to refresh its model picker.

AgentPort 3.1.29

AgentPort 3.1.29

Fixes the clipped update banner. Update progress, the ready message and Install update control now appear below the title bar in the scrollable page content. Status and error messages take their natural height instead of occupying the fixed-height navigation row.

Verified downloading, ready, installer rejection, navigation and no-update states, including desktop and narrow window layouts. Model selections, credentials, routing and context settings are unchanged.

Usage now lists every saved Codex subscription with its own reported allowance windows and reset times. Account management and performance/request details are expandable sections in Usage, replacing two sidebar tabs. Unavailable readings remain explicit; account allowances are never added together.

Available for Windows x64 and Apple Silicon macOS. The Mac preview supports Codex and API providers; local inference is not included. It is ad hoc signed and is not Apple-notarised.

AgentPort 3.1.28

AgentPort 3.1.28

API Models now shows only connected providers, grouped into collapsible lists. Search across those providers, filter to models in Codex or free models, and reveal more rows as needed. The separate Add models dialog is gone: switch on a catalogue model directly in its provider list. Existing model visibility choices are preserved.

Adds guided NVIDIA NIM setup in API Models. Open NVIDIA Build to create a key, enter it securely in AgentPort, browse the live NVIDIA catalogue and choose models for Codex. These models run on NVIDIA infrastructure. No local model download or local GPU is required.

NVIDIA provides free developer access for prototyping subject to its account limits and model availability. Models listed by the catalogue are not automatically verified for Codex tool use. Embedding, reranking, reward and document-parsing services are excluded from the chat list.

Connecting preserves model visibility choices. Select the models you want, then fully quit and reopen Codex to refresh its picker.

The OpenRouter setup flow remains available. Native GPT model settings are unchanged. Discovery and setup were checked with offline fixtures and the public NVIDIA catalogue; no authenticated inference request was sent.

Windows x64 release. Apple Silicon remains on 3.1.24.

AgentPort 3.1.27

AgentPort 3.1.27

Adds guided OpenRouter setup to API Models. Create a key from the direct link, save it securely through AgentPort, browse the current OpenRouter model catalogue, and choose models for Codex. The setup explains OpenRouter billing and the Codex restart needed to refresh its model picker.

Free routes remain subject to OpenRouter limits. Paid models use OpenRouter credits and account controls.

Windows x64 installer. Apple Silicon remains on 3.1.24.

AgentPort 3.1.26: Haiku tool-loading and token safeguards

AgentPort 3.1.26

Fixes oversized Haiku tool prompts. When Codex supplies its Code Mode execution tool, AgentPort keeps app and MCP definitions behind that client-owned mechanism instead of repeating every schema in the provider request. Core coding tools, collaboration, tools referenced by the conversation and explicit tool choices remain available. The router does not execute tools itself.

Older clients that omit the execution tool no longer trigger injection of the static app catalogue. If their remaining tool definitions exceed 256 KB, AgentPort refuses the request locally with a clear message. It does not silently drop tools or send the oversized request to Haiku. Restart Codex to refresh its model capabilities if this warning appears.

This safeguard applies to the Haiku Code Mode route. Native GPT models, their context windows, provider token accounting, and other model routes are unchanged. Existing allowance consumption cannot be undone.

Offline regression checks cover 1,400 MCP tool definitions being kept behind Codex execution, preservation of required tools, native routing, and an oversized request returning HTTP 400 without reaching the mock provider. No paid Haiku request was sent to test this release. Actual quota savings still require confirmation from a subsequent user request.

Windows x64 release. The Apple Silicon download remains on 3.1.24 pending a working Mac build environment.

AgentPort 3.1.25: subscription switching and automatic update downloads

AgentPort 3.1.25

Choose Use this subscription in Codex Accounts to send new native Codex turns through a saved ChatGPT subscription. You can use two signed-in accounts in one Codex instance. Each keeps its own allowance; this does not merge or increase either account's limits. Existing turns, retries and subagents keep their bound account. Explicit per-chat selections in Context Manager still take precedence. Choose Codex app login (default) to return to the desktop's own account.

Successful browser sign-in no longer reports a false failure when status recovery finishes before the desktop callback. Overlapping account sign-ins are blocked with a clear explanation. A saved login that becomes usable also clears its stale error notice.

The app checks GitHub for a newer stable release on launch and downloads the matching installer automatically. Published size and SHA-256 must match before installation is offered. Windows installs from the Install update button after active work has finished. On Apple Silicon, this opens the verified DMG for the normal drag-to-Applications installation. The Mac build remains an ad hoc-signed preview, not Apple-notarised.

Provider catalogue discovery already refreshes on every launch, including connected OpenCode Go/Zen sources and enabled free routes. Regression checks confirm a fresh launch bypasses the cache, offline use retains the saved list, and existing model choices are preserved. New models are available to choose in API Models; publication into Codex still requires its picker to refresh.

This release does not change native GPT context-window settings or API-provider credentials. Install the matching router and desktop update before using the subscription selector; an older running router is detected rather than silently ignoring the selection.

Verification includes two distinct saved subscriptions each returning a completed live response through AgentPort, account-routing regression tests, sign-in recovery checks, renderer interaction tests, and updater checksum/rejection tests. Automatic installation of a future release cannot be verified until a newer compatible release exists.

Delivery: Windows x64 is available. Apple Silicon remains on 3.1.24 because GitHub build jobs are blocked by account billing and the connected Mac did not respond. No 3.1.25 Mac installer has been published.

AgentPort 3.1.24 - Codex Desktop sign-in fix

AgentPort 3.1.24

Fix account sign-in with Codex Desktop

The Codex Accounts page could report Codex CLI is not installed when the desktop app was installed but no separate CLI was on PATH. This was an AgentPort detection bug.

Account sign-in now uses the shared desktop-aware Codex detector. It can use the helper already bundled with the desktop app rather than requiring a separate global CLI installation. The detector also recognises the newer ChatGPT Mac app's nested codex-cli layout alongside the older Codex app layout and Windows version-hashed installations.

When no helper is available, the message points to installing or updating the desktop app instead of presenting a CLI-only error. The existing per-account home, browser sign-in, process ownership, retry and finalisation safeguards are unchanged. Adding an account does not replace your main desktop login.

Update and retry

Finish active tasks and quit AgentPort. Install this update over your existing copy, then reopen AgentPort. In Codex Accounts, click Sign in beside the account you already added. There is no need to remove it or create it again. Complete sign-in in your browser, then choose the account for a conversation in Context Manager.

You can continue using the Codex desktop app. No manual terminal login, global CLI install or API key is required for this setup path when a supported desktop helper is present.

Verification and limits

Regression coverage includes desktop-only detection, current and legacy Mac layouts, Windows version-hashed helper selection, packaged-helper lookup, missing-helper recovery instructions and browser handoff into an isolated account home. Release verification checks the packaged IPC and resolver against the tested source.

Real account authentication still requires you to complete browser sign-in. Automated tests do not claim a successful live cross-account conversation or acceptance of account-scoped reasoning and uploaded files.

The Apple Silicon build remains an ad hoc-signed preview, not Apple-notarised, with the same API-only scope. Local inference is not included in the Mac preview.

AgentPort 3.1.23 - Codex Accounts

AgentPort 3.1.23

Add Codex accounts in one place

Codex Accounts is now a visible item in the left sidebar. Click Add Codex account, complete sign-in in your browser, and repeat for each ChatGPT subscription you own. An optional name such as Personal or Work makes accounts easier to recognise.

Choose Choose account for a chat to open Context Manager. Select an account on an existing Codex conversation and click Apply. Different conversations can use different accounts at the same time, without creating separate Codex windows.

A running task keeps its account. Changes apply at a confirmed new Codex turn, including after tool calls or a router restart. The existing conversation and its Open app/Resume session identity are preserved.

Clearer setup and recovery

• Duplicate additions are blocked while sign-in is starting.
• If the browser cannot open, the saved account stays in the list. Use Sign in on that row to retry.
• Use for app is separate: it changes the main Codex desktop login after Codex closes. Per-chat choices stay in Context Manager.
• Existing API provider routes keep their own credentials.
• The release source retains executable permissions for its Mac/Linux launchers.

Install the update

After active tasks finish, fully quit AgentPort and Codex. Run the Windows installer over your current installation, or replace AgentPort in Applications using the Apple Silicon DMG. Reopen AgentPort and Codex.

Then open Codex Accounts → Add Codex account. No Git checkout, terminal build or separate account-manager installation is required.

The installer contains matching versions of the interface and router. Release checks verify account-routing behaviour, the production interface, platform packages and installer checksums.

Compatibility notes

Cross-account acceptance of existing account-scoped reasoning, compaction state and uploaded files has not been tested against real accounts. The new account must also support the chosen model. Backend refusals remain visible rather than deleting context or silently using another account. Multiple accounts do not make a single response faster.

The Mac package is an Apple Silicon preview. It is ad hoc signed, not Apple-notarised, and retains the existing API-only scope. Local inference is not included. See the Mac installation notes for the existing installation guidance.

Application source and development history remain private. This public repository distributes installers, checksums, release notes and the download page only.

AgentPort 3.1.22 - clear model selection and free model groups

AgentPort 3.1.22 Apple Silicon preview

Connecting or enabling a provider now prompts you to choose models. The API
Models page explains that connecting an account does not automatically show
every model, and that Codex needs a full quit and reopen after model selections
change. Existing visible and hidden model choices are preserved.

API Models now groups models with a free provider route together in a clearly
labelled Free models section. Paid and subscription models appear separately.
Disconnected free routes stay in the free group and explain that a provider
connection is needed. For models with several routes, expand the model to
choose its free provider. Provider usage limits can still apply.

The source is maintained privately in
OA-Motion/project_development_files/AgentPort. The public AgentPort repository
contains release downloads and product documentation only.

Verification and remaining limitations

• On the M5 Mac, a fresh Codex app-server returned all 41 selected OpenCode models,
  including Claude Haiku 5.5 and DeepSeek V4.1 Flash.
• A real DeepSeek Flash request reached OpenCode, which returned HTTP 429:
  Go usage limit exceeded. Successful inference remains unverified on that account.
• The installed 3.1.21 app passed signature integrity verification but was
  rejected by Gatekeeper. This update does not change the signing identity.
  The Mac app remains ad hoc signed and is not notarised. Open Anyway alone
  did not resolve the reported launch failure. See docs/MACOS-DESKTOP.md in
  the private source for the investigation and reported workaround.
• Local models remain outside the Mac preview's scope.

AgentPort 3.1.21 - Apple Silicon preview

AgentPort 3.1.21: Apple Silicon preview

This release adds a standalone Mac installer for Apple Silicon, including M5.
It focuses on Codex and API providers such as OpenCode Go/Zen. Local model
downloads and inference are not included in this Mac release. Native GPT model
metadata, context settings and ChatGPT authentication are unchanged.

Install on your Mac

1. Download agentport-3.1.21-macos-arm64.dmg, open it and drag AgentPort to Applications.
2. Open AgentPort from Applications. This preview is ad-hoc signed, rather than
   Apple-notarised. If macOS blocks it, use System Settings > Privacy & Security >
   Open Anyway after attempting to open the app. Do not disable Gatekeeper.
3. In API Models, connect your OpenCode account/key or another provider and select models.
4. In Harness, turn on Codex. The first connection installs the pinned Python
   gateway dependencies automatically. Internet access is required; allow several
   minutes for this step. Node and uv are already bundled, so Homebrew is unnecessary.
5. Fully quit and reopen Codex, start a new chat and select the routed model.

The router runs as your user's launchd service after setup. AgentPort has its own
menu-bar icon; it does not install the older separate Swift companion. To update,
quit AgentPort and replace the app in Applications with the newer Mac installer.
The owned router checkout is refreshed while saved settings and environments stay
in place. Router files live in ~/.local/share/agentport, bundled tools in
~/.local/share/agentport-runtime, and state in ~/.codex/agentport.

The stable Windows release remains 3.1.20. This Mac build is published as a preview
until installation and real provider requests are confirmed on a user's Mac.

Verification

The Mac build pipeline verifies the ad-hoc app signature, arm64 runtime execution,
npm links, provider discovery, launchd rendering and a simulated owned-checkout
upgrade. These package checks do not replace a live Codex/provider test on the M5.