Factory CLI with CLIProxyAPI for Claude and ChatGPT Subscription Use

FACTORY_CLIProxyAPI_Claude_ChatGPT.md

Using Factory CLI (Droid) with CLIProxyAPI for Custom Models

---

⚠️ IMPORTANT DISCLAIMER

This guide describes a proof-of-concept configuration for educational and experimental purposes only.

By using this method, you acknowledge and accept the following:

• Terms of Service Risk: This approach may violate the Terms of Service of AI model providers (Anthropic, OpenAI, etc.). You are solely responsible for ensuring compliance with all applicable terms and policies.

• Account Risk: Model providers may detect this usage pattern and take punitive action, including but not limited to account suspension, permanent ban, or loss of access to paid subscriptions.

• No Guarantees: Providers may change their APIs, authentication mechanisms, or policies at any time, rendering this method inoperable without notice.

• Assumption of Risk: By proceeding, you assume all legal, financial, and technical risks. The authors and contributors of this guide and CLIProxyAPI bear no responsibility for any consequences arising from your use of this method.

Use at your own risk. Proceed only if you understand and accept these risks.

---

Executive Summary

This guide explains how to run Factory's Droid CLI against personal subscriptions instead of pay-per-token API keys by placing CLIProxyAPI between Factory and the upstream providers. Two flows are covered:

1. Claude Code Max (Anthropic OAuth) → exposes Anthropic's Messages API through provider: "anthropic"
2. ChatGPT Plus/Pro (OpenAI OAuth) → exposes GPT‑5 / GPT‑5 Codex via provider: "openai" (Responses API)

CLIProxyAPI swaps the dummy API key that Factory sends for the real OAuth bearer token, auto-refreshes tokens, and keeps request/response bodies in the provider's native format.

Architecture Overview

Factory CLI  →  [Anthropic format + API key]  →
                      CLIProxyAPI  →  [Anthropic format + OAuth]      →  Anthropic API

Factory CLI  →  [OpenAI Responses schema + API key]  →
                      CLIProxyAPI  →  [OpenAI Responses schema + OAuth]  →  ChatGPT Codex API

Prerequisites

• Linux/macOS/WSL2 host
• Go 1.24+ and Git
• Active Claude Code Max (or Claude Pro) subscription for Anthropic login
• Active ChatGPT Plus/Pro subscription for OpenAI Codex login
• Factory CLI installed (curl -fsSL https://app.factory.ai/cli | sh)

Open outbound HTTPS to api.anthropic.com and chatgpt.com (CLIProxyAPI defaults to port 8317 locally, OAuth callbacks use 54545 for Anthropic and 1455 for OpenAI).

Step 1: Install Go (if needed)

wget https://go.dev/dl/go1.25.1.linux-amd64.tar.gz
sudo rm -rf /usr/local/go
sudo tar -C /usr/local -xzf go1.25.1.linux-amd64.tar.gz
export PATH=/usr/local/go/bin:$PATH
export GOPATH=$HOME/go
export PATH=$GOPATH/bin:$PATH
go version

Step 2: Clone & Build CLIProxyAPI

Note: Claude 4.5 Sonnet support requires v6.0.15+ of CLIProxyAPI.

git clone https://github.com/luispater/CLIProxyAPI.git
cd CLIProxyAPI
go build -o cli-proxy-api ./cmd/server

Temporary patch for ChatGPT Responses

Note: The step 1 code change below is fixed and no longer required in v6.0.10+ of CLIProxyAPI

Until the upstream project merges the Codex compatibility fix, remove the unsupported token-limit fields from the Codex Responses translator:

1. Edit internal/translator/codex/openai/responses/codex_openai-responses_request.go and delete both max_output_tokens and max_completion_tokens before forwarding:

   // Codex Responses rejects token limit fields, so remove them before forwarding.
   rawJSON, _ = sjson.DeleteBytes(rawJSON, "max_output_tokens")
   rawJSON, _ = sjson.DeleteBytes(rawJSON, "max_completion_tokens")

2. Run gofmt on the file:

   gofmt -w internal/translator/codex/openai/responses/codex_openai-responses_request.go

3. Rebuild the binary:

   go build -o cli-proxy-api ./cmd/server

This mirrors the fix that prevents OpenAI from rejecting requests with {"detail":"Unsupported parameter: max_output_tokens"}.

Step 3: Configure the proxy (config.yaml)

port: 8317
remote-management:
  allow-remote: false
  secret-key: ""
auth-dir: "~/.cli-proxy-api"
debug: false
logging-to-file: false
usage-statistics-enabled: true
proxy-url: ""
request-retry: 3
quota-exceeded:
  switch-project: true
  switch-preview-model: true
auth:
  providers: []
generative-language-api-key: []

Optional runtime toggles

• Set debug: true to enable extra logging
• Set request-log: true to capture full request/response bodies under logs/

Step 4: OAuth Authentication Setup

4.1 Claude Code Max / Claude Pro

./cli-proxy-api --claude-login

• Launches browser on port 54545 to complete authentication
• Saves token file at ~/.cli-proxy-api/claude-<email>.json

4.2 ChatGPT Plus/Pro (Codex)

./cli-proxy-api --codex-login

• Launches browser on port 1455 to finish ChatGPT OAuth
• Stores token at ~/.cli-proxy-api/codex-<email>.json

Headless / remote tip

When running on a server without a browser:

1. Start the login command (--claude-login or --codex-login). Copy the printed authorization URL.
2. Paste the URL into a browser on your local machine.
3. After approving access, the browser fails to reach the callback (because the user is remote) and shows a http://127.0.0.1:<port>/callback?... URL.
4. Copy that full failed redirect URL.
5. In another SSH session to the headless server, run curl '<copied_url>' to complete the flow.

This avoids SSH port forwarding while still delivering the OAuth code to the proxy.

Step 5: Configure Factory CLI (~/.factory/config.json)

{
  "custom_models": [
    {
      "model": "claude-sonnet-4-5-20250929",
      "base_url": "http://localhost:8317",
      "api_key": "dummy-not-used",
      "provider": "anthropic"
    },
    {
      "model": "claude-opus-4-1-20250805",
      "base_url": "http://localhost:8317",
      "api_key": "dummy-not-used",
      "provider": "anthropic"
    },
    {
      "model": "claude-sonnet-4-20250514",
      "base_url": "http://localhost:8317",
      "api_key": "dummy-not-used",
      "provider": "anthropic"
    },
    {
      "model": "gpt-5",
      "base_url": "http://localhost:8317/v1",
      "api_key": "dummy-not-used",
      "provider": "openai"
    },
    {
      "model": "gpt-5-minimal",
      "base_url": "http://localhost:8317/v1",
      "api_key": "dummy-not-used",
      "provider": "openai"
    },
    {
      "model": "gpt-5-low",
      "base_url": "http://localhost:8317/v1",
      "api_key": "dummy-not-used",
      "provider": "openai"
    },
    {
      "model": "gpt-5-medium",
      "base_url": "http://localhost:8317/v1",
      "api_key": "dummy-not-used",
      "provider": "openai"
    },
    {
      "model": "gpt-5-high",
      "base_url": "http://localhost:8317/v1",
      "api_key": "dummy-not-used",
      "provider": "openai"
    },
    {
      "model": "gpt-5-codex",
      "base_url": "http://localhost:8317/v1",
      "api_key": "dummy-not-used",
      "provider": "openai"
    },
    {
      "model": "gpt-5-codex-low",
      "base_url": "http://localhost:8317/v1",
      "api_key": "dummy-not-used",
      "provider": "openai"
    },
    {
      "model": "gpt-5-codex-medium",
      "base_url": "http://localhost:8317/v1",
      "api_key": "dummy-not-used",
      "provider": "openai"
    },
    {
      "model": "gpt-5-codex-high",
      "base_url": "http://localhost:8317/v1",
      "api_key": "dummy-not-used",
      "provider": "openai"
    }
  ]
}

Why /v1? Factory appends /messages automatically for provider: "anthropic", so the proxy sees /v1/messages even when the base URL ends at :8317. For provider: "openai", Factory appends only /responses, so we add /v1 to the base URL to match CLIProxyAPI's routing.

The Codex executor in CLIProxyAPI understands the full GPT‑5 family above. The "minimal"/"low"/"medium"/"high" variants all fan in to the gpt-5 backend with different reasoning.effort settings; the gpt-5-codex-* variants do the same on the Codex endpoint. Additional aliases such as codex-mini-latest are auto-registered but are optional.

Step 6: Run the proxy

./cli-proxy-api --config config.yaml

Or run in background (nohup), or as a systemd service (see below). The startup log should mention "server clients and configuration updated: …".

Step 7: Use Factory CLI

droid
/model   # pick claude-sonnet-4-5-20250929, claude-opus-4-1-20250805, gpt-5, etc.

Factory now calls the proxy with the dummy key; CLIProxyAPI injects the OAuth bearer token and relays responses.

Production options

systemd service

Create /etc/systemd/system/cli-proxy-api.service (adjust paths):

[Unit]
Description=CLI Proxy API bridge for Factory
After=network.target

[Service]
Type=simple
User=your-user
WorkingDirectory=/path/to/CLIProxyAPI
ExecStart=/path/to/CLIProxyAPI/cli-proxy-api --config config.yaml
Restart=on-failure
RestartSec=10

[Install]
WantedBy=multi-user.target

Then:

sudo systemctl daemon-reload
sudo systemctl enable cli-proxy-api
sudo systemctl start cli-proxy-api
sudo systemctl status cli-proxy-api

Docker

FROM golang:1.25-alpine AS builder
WORKDIR /app
COPY . .
RUN go build -o cli-proxy-api ./cmd/server

FROM alpine:latest
RUN apk --no-cache add ca-certificates
WORKDIR /root
COPY --from=builder /app/cli-proxy-api .
COPY config.yaml .
CMD ["./cli-proxy-api", "--config", "config.yaml"]

Build and run:

docker build -t cli-proxy-api .
docker run -d -p 8317:8317 \
  -v ~/.cli-proxy-api:/root/.cli-proxy-api \
  -v $(pwd)/config.yaml:/root/config.yaml \
  cli-proxy-api

Troubleshooting

Symptom	Likely cause	Fix
404 /responses	Missing /v1 on OpenAI base URL	Use http://localhost:8317/v1
401/403 responses	OAuth token expired	Re-run --claude-login / --codex-login
A.match is not a function in Factory	Client bug parsing tool call	Provide logs to Factory (proxy log captures SSE stream)
listen tcp :8317: bind	Port already used	Stop existing process or change port

Enable debug: true for verbose logging, request-log: true to capture full payloads in logs/ for audits.

Security Checklist

• Token files in ~/.cli-proxy-api are 0600; never commit them
• Proxy binds to localhost only
• All upstream traffic is HTTPS
• Auto-refresh keeps tokens current (watch logs for refresh errors)

Supported models snapshot

• Claude: claude-sonnet-4-5-20250929 (Claude 4.5 Sonnet), claude-opus-4-1-20250805, claude-sonnet-4-20250514 (and others loaded from registry)
• OpenAI: gpt-5, gpt-5-codex, gpt-5-… family, codex-mini-latest

Verification checklist

1. ./cli-proxy-api --config config.yaml
2. droid → /model → select custom model
3. Run a quick command (e.g. "what day is it?") and verify proxy logs show 200 responses

References

• CLIProxyAPI repo: https://github.com/luispater/CLIProxyAPI
• Factory CLI docs: https://docs.factory.ai/cli
• Anthropic docs: https://docs.anthropic.com
• ChatGPT Codex flow (same OAuth used by codex-cli)

Thanks for that! I've vibe-coded a macOS menubar utility that manages all of that without having to build anything
https://github.com/automazeio/vibeproxy