Factory CLI with Claude Subscription / ChatGPT Codex via CLIProxyAPI

FACTORY_PROXY_CC.md

Executive Summary

This guide documents how to use Factory's Droid CLI with your Claude Code Max subscription (OAuth authentication) instead of pay-per-token API keys. The solution leverages CLIProxyAPI as a transparent authentication proxy that converts API key requests from Factory CLI into OAuth-authenticated requests for Anthropic's API.

Architecture Overview

Factory CLI → [Anthropic Format + API Key] → CLIProxyAPI → [Anthropic Format + OAuth] → Anthropic API
                                                  ↓
                                          (Auth Header Swap)

Key Components

1. Factory Droid CLI: AI-powered development assistant with BYOK (Bring Your Own Key) support
2. CLIProxyAPI: Open-source proxy server that handles OAuth authentication for CLI models
3. Claude Code Max Subscription: Consumer subscription using OAuth tokens instead of API keys
4. Anthropic Messages API: Native API endpoint for Claude models

Why This Solution?

The Problem

• Factory CLI expects API keys (x-api-key header) for Anthropic integration
• Claude Code Max subscriptions use OAuth tokens (Authorization: Bearer)
• Anthropic's public API doesn't accept OAuth tokens directly
• No need for expensive pay-per-token API access when you have a Max subscription

The Solution

CLIProxyAPI acts as a transparent proxy that:

• Accepts native Anthropic format requests from Factory CLI
• Replaces dummy API keys with valid OAuth tokens
• Handles OAuth token refresh automatically
• Preserves request/response format (no translation needed)

Prerequisites

System Requirements

• Linux, macOS, or WSL2 on Windows
• Go 1.24 or higher (for building from source)
• Git for cloning the repository
• Active Claude Code Max subscription
• Factory CLI installed and configured

Network Requirements

• Port 8317 (default) or any available port for the proxy
• Port 54545 for OAuth callback during initial setup
• Internet access to Anthropic's API endpoints

Installation Guide

Step 0: Install Factory Droid

• Sign up at https://factory.ai/ and install the bridge
• Install the Droid CLI.

curl -fsSL https://app.factory.ai/cli | sh

Step 1: Install Go (if not already installed)

# Download and install Go
brew install go

# Add to PATH (add to ~/.bashrc or ~/.zshrc for persistence)
export PATH=$PATH:/usr/local/go/bin
export GOPATH=$HOME/go
export PATH=$PATH:$GOPATH/bin

# Verify installation
go version

Step 2: Clone and Build CLIProxyAPI

# Clone the repository
git clone https://github.com/luispater/CLIProxyAPI.git
cd CLIProxyAPI

# Build the binary
go build -o cli-proxy-api ./cmd/server

# Verify the binary was created
ls -la cli-proxy-api

Step 3: Configure the Proxy

Create a config.yaml file in the CLIProxyAPI directory:

# Server port (use any available port)
port: 8317

# Management API settings
remote-management:
  allow-remote: false
  secret-key: ""  # Leave empty to disable management API

# Authentication directory (stores OAuth tokens)
auth-dir: "~/.cli-proxy-api"

# Logging configuration
debug: false
logging-to-file: false

# Usage statistics
usage-statistics-enabled: true

# Network proxy (if needed for corporate networks)
proxy-url: ""

# Retry configuration
request-retry: 3

# Quota management
quota-exceeded:
  switch-project: true
  switch-preview-model: true

# Disable request authentication (Factory handles its own)
auth:
  providers: []

# No API keys needed for Claude OAuth
generative-language-api-key: []

Step 4: OAuth Authentication Setup

Run the OAuth login flow to authenticate with your Claude Code account:

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

For getting OpenAI Codex working with this:

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

This will:

1. Start a local OAuth callback server on port 54545
2. Provide an authentication URL (open in browser if not automatic)
3. Complete the OAuth flow with Anthropic / OpenAI
4. Save tokens to ~/.cli-proxy-api/claude-{email}.json

For Remote Servers: If running on a remote server, you'll need an SSH tunnel:

# On your local machine (not the server):
ssh -L 54545:127.0.0.1:54545 user@your-server-ip

Step 5: Configure Factory CLI

Create or modify ~/.factory/config.json:

{
  "custom_models": [
    {
      "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",
      "api_key": "dummy-not-used",
      "provider": "openai"
    },
    {
      "model": "gpt-5-minimal",
      "base_url": "http://localhost:8317",
      "api_key": "dummy-not-used",
      "provider": "openai"
    },
    {
      "model": "gpt-5-medium",
      "base_url": "http://localhost:8317",
      "api_key": "dummy-not-used",
      "provider": "openai"
    },
    {
      "model": "gpt-5-high",
      "base_url": "http://localhost:8317",
      "api_key": "dummy-not-used",
      "provider": "openai"
    },
    {
      "model": "gpt-5-codex",
      "base_url": "http://localhost:8317",
      "api_key": "dummy-not-used",
      "provider": "openai"
    },
    {
      "model": "gpt-5-codex-high",
      "base_url": "http://localhost:8317",
      "api_key": "dummy-not-used",
      "provider": "openai"
    }
  ]
}

Note: The api_key field is required by Factory but ignored by the proxy (OAuth is used instead).

Running the System

Start the Proxy Server

# Start in foreground (for testing)
./cli-proxy-api --config config.yaml

# Or start in background
nohup ./cli-proxy-api --config config.yaml > proxy.log 2>&1 &

# Or use systemd service (see below)

Expected output:

CLIProxyAPI Version: dev, Commit: none, BuiltAt: unknown
API server started successfully
server clients and configuration updated: 1 clients (1 auth files)

Using Factory CLI

1. Start Factory Droid:

   droid

2. Select your custom model:

   /model
   

   Choose either:

   • claude-opus-4-1-20250805 (Claude Opus 4.1)
   • claude-sonnet-4-20250514 (Claude Sonnet 4)
   • or other openAI models.

3. Use as normal - Factory will now use your Claude Code subscription!

Production Setup

Systemd Service (Linux)

Create /etc/systemd/system/cli-proxy-api.service:

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

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

[Install]
WantedBy=multi-user.target

Enable and start:

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

Docker Deployment

Create a Dockerfile:

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 cli-proxy-api

Troubleshooting

Common Issues and Solutions

Port Already in Use

Error: listen tcp :8317: bind: address already in use

Solution: Change the port in both config.yaml and ~/.factory/config.json.

OAuth Token Expired

Error: OAuth token unavailable or expired

Solution: Re-run ./cli-proxy-api --claude-login to refresh authentication.

Factory Can't Connect to Proxy

Error: Connection refused

Solutions:

• Ensure proxy is running: ps aux | grep cli-proxy-api
• Check correct port in Factory config matches proxy config
• Verify firewall allows localhost connections

Model Not Found

Error: Model claude-xxx not found

Solution: Use exact model names from the supported list:

• claude-opus-4-1-20250805
• claude-sonnet-4-20250514

Debug Mode

Enable debug logging for troubleshooting:

# In config.yaml
debug: true
logging-to-file: true

Check logs in logs/ directory relative to config file.

Security Considerations

OAuth Token Security

• Tokens stored in ~/.cli-proxy-api/ with user-only permissions
• Never commit token files to version control
• Tokens auto-refresh before expiration
• No tokens appear in logs (automatically redacted)

Network Security

• Proxy binds to localhost only by default
• No external access without explicit configuration
• HTTPS used for all Anthropic API communication
• Optional proxy authentication can be enabled

Best Practices

1. Run proxy as non-root user
2. Use systemd service with restart policies
3. Monitor logs for authentication failures
4. Rotate OAuth tokens periodically (re-login)
5. Keep CLIProxyAPI updated for security patches

Supported Models

Currently Available

• Claude Opus 4.1 (claude-opus-4-1-20250805)

  • Most capable model for complex reasoning
  • Best for architecture, planning, deep analysis

• Claude Sonnet 4 (claude-sonnet-4-20250514)

  • Balanced performance and speed
  • Ideal for general development tasks

Adding New Models

When Anthropic releases new models, update Factory config with the exact model ID:

{
  "model": "claude-{new-model-id}",
  "base_url": "http://localhost:8317",
  "api_key": "dummy-not-used",
  "provider": "anthropic"
}

Technical Details

How the Proxy Works

1. Request Flow:

   • Factory sends Anthropic-format request with dummy API key
   • Proxy receives request on /v1/messages endpoint
   • Proxy strips X-Api-Key header
   • Proxy adds Authorization: Bearer {oauth_token}
   • Proxy adds required beta headers for OAuth
   • Request forwarded to Anthropic unchanged

2. Response Flow:

   • Anthropic responds in native format
   • Proxy streams response back to Factory
   • No format translation needed

3. Token Management:

   • OAuth tokens cached until near expiration
   • Automatic refresh using refresh token
   • Multiple account support with round-robin

Key Files and Directories

~/
├── .factory/
│   └── config.json          # Factory CLI configuration
├── .cli-proxy-api/
│   └── claude-*.json        # OAuth tokens (auto-generated)
└── projects/CLIProxyAPI/
    ├── cli-proxy-api        # Compiled binary
    ├── config.yaml          # Proxy configuration
    └── logs/               # Log files (if enabled)

Performance Optimization

Latency Reduction

• No format translation overhead (native Anthropic format)
• Direct pass-through of streaming responses
• Connection pooling for API requests
• Local caching of OAuth tokens

Resource Usage

• Minimal CPU usage (< 1% idle, < 5% active)
• Low memory footprint (< 50MB typical)
• No disk I/O except for logs (if enabled)
• Stateless operation (except OAuth tokens)

Monitoring and Maintenance

Health Checks

# Check if proxy is running
curl -s http://localhost:8317/health || echo "Proxy not responding"

# View real-time logs
tail -f logs/app.log  # If file logging enabled

# Check OAuth token status
ls -la ~/.cli-proxy-api/

Regular Maintenance

• Weekly: Check logs for errors
• Monthly: Update CLIProxyAPI if new version available
• Quarterly: Re-authenticate OAuth (preventive)
• As Needed: Adjust port if conflicts arise

Conclusion

This solution enables seamless integration between Factory's Droid CLI and Claude Code Max subscriptions, eliminating the need for pay-per-token API access. The CLIProxyAPI serves as a lightweight, transparent proxy that handles OAuth authentication while preserving the native Anthropic API format that Factory already supports.

Benefits Achieved

• ✅ Use Claude Code Max subscription with Factory CLI
• ✅ No API key costs
• ✅ Automatic token refresh
• ✅ Native Anthropic format (no translation overhead)
• ✅ Simple one-time setup
• ✅ Production-ready with systemd/Docker support

Next Steps

1. Set up the proxy following this guide
2. Test with a simple Factory CLI interaction
3. Configure for production use if successful
4. Share with team members who have Claude Code subscriptions

Support and Resources

CLIProxyAPI Resources

• GitHub Repository: https://github.com/luispater/CLIProxyAPI
• Issues/Support: https://github.com/luispater/CLIProxyAPI/issues
• Documentation: See repository README.md

Factory CLI Resources

• Documentation: https://docs.factory.ai/cli
• BYOK Configuration: https://docs.factory.ai/cli/configuration/byok
• Support: Contact Factory support team

Anthropic/Claude Resources

• API Documentation: https://docs.anthropic.com
• Model Information: https://docs.anthropic.com/en/docs/about-claude/models
• Claude Code: https://claude.ai

Appendix: Quick Reference

Essential Commands

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

# OAuth setup
./cli-proxy-api --claude-login

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

# Test with curl
curl -X POST http://localhost:8317/v1/messages \
  -H "Content-Type: application/json" \
  -H "x-api-key: dummy" \
  -H "anthropic-version: 2023-06-01" \
  -d '{"model":"claude-opus-4-1-20250805","messages":[{"role":"user","content":"Hello"}],"max_tokens":100}'

# Use with Factory
droid
/model  # Select custom model

Configuration Templates

All configuration templates are provided in the main guide above and can be copied directly.

---

Hmmm I keep getting 404 for any of the Codex models? Verified successful Codex Authentication. Anyone else run into these problems?

Hmmm I keep getting 404 for any of the Codex models? Verified successful Codex Authentication. Anyone else run into these problems?

yup

Everytime it goes to compact in claude code proxy it fails and i have to create a new conversation

x2 same here

Everytime it goes to compact in claude code proxy it fails and i have to create a new conversation

it works using the openai api proxy but it breaks tool calling

{
  "custom_models": [
    {
      "model": "claude-sonnet-4-5-20250929",
      "base_url": "http://localhost:8317/v1",
      "api_key": "dummy-not-used",
      "provider": "openai"
    }
  ]
}

PD: I've found that clewdr works much better with factory cli https://github.com/Xerxes-2/clewdr

@ildunari and @Designedforusers you need to do as @ultramathi stated. Go to Step 5, where you're configuring the the Factory CLI. Update the base url: "base_url": "http://localhost:8317/v1" (places a /v1 at the end of the URL). save the config file, and from there you'll be fine.

Thank you for the help! got it working. Another issue though, I’m getting empty returns when I try to run sub droids/tasks? Is that to be expected with custom models we’re running? As in, can we only use sub agents if we use factory and also don’t have any custom models listed in the config or what’s a way to get that to work as well?

Thank you for the help! got it working. Another issue though, I’m getting empty returns when I try to run sub droids/tasks? Is that to be expected with custom models we’re running? As in, can we only use sub agents if we use factory and also don’t have any custom models listed in the config or what’s a way to get that to work as well?

You'll have to forgive me mate, but I don't know. I haven't played around with it... All I've done is get it connected then head to the gym. 😂

Ok, going to lift is an acceptable excuse. What r ur lifts and stats? 😂

Ha! Enough to be called a liar for sharing on the internet. 😂 Have a good one bud! Get Outlook for Android<https://aka.ms/AAb9ysg>

…

________________________________ From: ildunari ***@***.***> Sent: Saturday, October 4, 2025 10:13:54 AM To: ildunari ***@***.***> Cc: Comment ***@***.***> Subject: Re: chandika/FACTORY_PROXY_CC.md @ildunari commented on this gist.

________________________________ Ok, going to lift is an acceptable excuse. What r ur lifts and stats? 😂 — Reply to this email directly, view it on GitHub<https://gist.github.com/chandika/c4b64c5b8f5e29f6112021d46c159fdd#gistcomment-5786600> or unsubscribe<https://github.com/notifications/unsubscribe-auth/BTQGHFCAASU2K4EOZPILVWD3V4UOFBFHORZGSZ3HMVZKMY3SMVQXIZNMON2WE2TFMN2F65DZOBS2WR3JON2EG33NNVSW45FGORXXA2LDOOIYFJDUPFYGLJDHNFZXJJLWMFWHKZNJGE2DCMRTGQ4TQNNKMF2HI4TJMJ2XIZLTSOBKK5TBNR2WLKBZGUYTQNJVG432I3TBNVS2QYLDORXXEX3JMSBKK5TBNR2WLJDUOJ2WLJDOMFWWLO3UNBZGKYLEL5YGC4TUNFRWS4DBNZ2F6YLDORUXM2LUPGBKK5TBNR2WLJDHNFZXJJDOMFWWLK3UNBZGKYLEL52HS4DF>. You are receiving this email because you commented on the thread. Triage notifications on the go with GitHub Mobile for iOS<https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android<https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>.

@ildunari and @Designedforusers you need to do as @ultramathi stated. Go to Step 5, where you're configuring the the Factory CLI. Update the base url: "base_url": "http://localhost:8317/v1" (places a /v1 at the end of the URL). save the config file, and from there you'll be fine.

yup got it working now, only issue is /compress doesnt work i tried fixing it no luck so far i think its a droid cli issue?

Everytime it goes to compact in claude code proxy it fails and i have to create a new conversation

it works using the openai api proxy but it breaks tool calling

{
  "custom_models": [
    {
      "model": "claude-sonnet-4-5-20250929",
      "base_url": "http://localhost:8317/v1",
      "api_key": "dummy-not-used",
      "provider": "openai"
    }
  ]
}

PD: I've found that clewdr works much better with factory cli https://github.com/Xerxes-2/clewdr

does clewdr allow for compaction and tool calling? seems like we have to decide on tradeoffs for each implementation

it's very eady to get 500 error ( for codex ) and no tool call usage for sonnet 4.5

I got a lot of 500 error (no body) anyone know why? using codex. Seem like when it try to compress then "Error: 500 status code (no body)"

Yeah, it works for claude opus and sonnet, but not for codex

Everytime it goes to compact in claude code proxy it fails and i have to create a new conversation

it works using the openai api proxy but it breaks tool calling

{
  "custom_models": [
    {
      "model": "claude-sonnet-4-5-20250929",
      "base_url": "http://localhost:8317/v1",
      "api_key": "dummy-not-used",
      "provider": "openai"
    }
  ]
}

PD: I've found that clewdr works much better with factory cli https://github.com/Xerxes-2/clewdr

does clewdr allow for compaction and tool calling? seems like we have to decide on tradeoffs for each implementation

yes, clewdr supports all the functionality

Can't get clewdr running with anthropic stuff

Everytime it goes to compact in claude code proxy it fails and i have to create a new conversation

it works using the openai api proxy but it breaks tool calling

{
  "custom_models": [
    {
      "model": "claude-sonnet-4-5-20250929",
      "base_url": "http://localhost:8317/v1",
      "api_key": "dummy-not-used",
      "provider": "openai"
    }
  ]
}

PD: I've found that clewdr works much better with factory cli https://github.com/Xerxes-2/clewdr

does clewdr allow for compaction and tool calling? seems like we have to decide on tradeoffs for each implementation

yes, clewdr supports all the functionality

i got the proxy working, but still no /compress working at all.

Does this break TOS for Claude MAX / OpenAI?

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

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

@ranaroussi Looks nice, you got compact error as well?

Yes. What I do, is switch the model to one of FactoryAI's built-in models, run /compact, and switch back to a the custom model.

Yes. What I do, is switch the model to one of FactoryAI's built-in models, run /compact, and switch back to a the custom model.

I see, thank you. I will try that out.

Switching to a model automatically compacts the conversation, that’s where I first encountered it

…

On Mon, Oct 6, 2025 at 4:23 AM mastertyko ***@***.***> wrote: ***@***.**** commented on this gist. ------------------------------ Yes. What I do, is switch the model to one of FactoryAI's built-in models, run /compact, and switch back to a the custom model. I see, thank you. I will try that out. — Reply to this email directly, view it on GitHub <https://gist.github.com/chandika/c4b64c5b8f5e29f6112021d46c159fdd#gistcomment-5788921> or unsubscribe <https://github.com/notifications/unsubscribe-auth/BO2MNNWC2JVI5EJQWTIPLFD3WIRINBFHORZGSZ3HMVZKMY3SMVQXIZNMON2WE2TFMN2F65DZOBS2WR3JON2EG33NNVSW45FGORXXA2LDOOIYFJDUPFYGLJDHNFZXJJLWMFWHKZNJGE2DCMRTGQ4TQNNKMF2HI4TJMJ2XIZLTSOBKK5TBNR2WLKBRGEZTCMJUG442I3TBNVS2QYLDORXXEX3JMSBKK5TBNR2WLJDUOJ2WLJDOMFWWLO3UNBZGKYLEL5YGC4TUNFRWS4DBNZ2F6YLDORUXM2LUPGBKK5TBNR2WLJDHNFZXJJDOMFWWLK3UNBZGKYLEL52HS4DF> . You are receiving this email because you commented on the thread. Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub> .

Switching to a model automatically compacts the conversation, that’s where I first encountered it
…
On Mon, Oct 6, 2025 at 4:23 AM mastertyko @.> wrote: @.* commented on this gist. ------------------------------ Yes. What I do, is switch the model to one of FactoryAI's built-in models, run /compact, and switch back to a the custom model. I see, thank you. I will try that out. — Reply to this email directly, view it on GitHub https://gist.github.com/chandika/c4b64c5b8f5e29f6112021d46c159fdd#gistcomment-5788921 or unsubscribe https://github.com/notifications/unsubscribe-auth/BO2MNNWC2JVI5EJQWTIPLFD3WIRINBFHORZGSZ3HMVZKMY3SMVQXIZNMON2WE2TFMN2F65DZOBS2WR3JON2EG33NNVSW45FGORXXA2LDOOIYFJDUPFYGLJDHNFZXJJLWMFWHKZNJGE2DCMRTGQ4TQNNKMF2HI4TJMJ2XIZLTSOBKK5TBNR2WLKBRGEZTCMJUG442I3TBNVS2QYLDORXXEX3JMSBKK5TBNR2WLJDUOJ2WLJDOMFWWLO3UNBZGKYLEL5YGC4TUNFRWS4DBNZ2F6YLDORUXM2LUPGBKK5TBNR2WLJDHNFZXJJDOMFWWLK3UNBZGKYLEL52HS4DF . You are receiving this email because you commented on the thread. Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub .

Nice find!!

just tried this, im getting "> Error: Failed to compress conversation for model switch"

…

On Mon, Oct 6, 2025 at 4:36 AM Ran Aroussi ***@***.***> wrote: ***@***.**** commented on this gist. ------------------------------ Switching to a model automatically compacts the conversation, that’s where I first encountered it … <#m_-8983328577459520007_> On Mon, Oct 6, 2025 at 4:23 AM mastertyko *@*.*> wrote: @.** commented on this gist. ------------------------------ Yes. What I do, is switch the model to one of FactoryAI's built-in models, run /compact, and switch back to a the custom model. I see, thank you. I will try that out. — Reply to this email directly, view it on GitHub https://gist.github.com/chandika/c4b64c5b8f5e29f6112021d46c159fdd#gistcomment-5788921 or unsubscribe https://github.com/notifications/unsubscribe-auth/BO2MNNWC2JVI5EJQWTIPLFD3WIRINBFHORZGSZ3HMVZKMY3SMVQXIZNMON2WE2TFMN2F65DZOBS2WR3JON2EG33NNVSW45FGORXXA2LDOOIYFJDUPFYGLJDHNFZXJJLWMFWHKZNJGE2DCMRTGQ4TQNNKMF2HI4TJMJ2XIZLTSOBKK5TBNR2WLKBRGEZTCMJUG442I3TBNVS2QYLDORXXEX3JMSBKK5TBNR2WLJDUOJ2WLJDOMFWWLO3UNBZGKYLEL5YGC4TUNFRWS4DBNZ2F6YLDORUXM2LUPGBKK5TBNR2WLJDHNFZXJJDOMFWWLK3UNBZGKYLEL52HS4DF . You are receiving this email because you commented on the thread. Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub . Nice find!! — Reply to this email directly, view it on GitHub <https://gist.github.com/chandika/c4b64c5b8f5e29f6112021d46c159fdd#gistcomment-5788935> or unsubscribe <https://github.com/notifications/unsubscribe-auth/BO2MNNTDQRGTQBT4L46LUNL3WISYHBFKMF2HI4TJMJ2XIZLTSOBKK5TBNR2WLJZRGE4DKNBVHCSG4YLNMWUGCY3UN5ZF62LEQKSXMYLMOVS2I5DSOVS2I3TBNVS3W5DIOJSWCZC7OBQXE5DJMNUXAYLOORPWCY3UNF3GS5DZQKSXMYLMOVS2IZ3JON2KI3TBNVS2W5DIOJSWCZC7OR4XAZNMON2WE2TFMN2F65DZOBS2WR3JON2EG33NNVSW45FGORXXA2LDOOIYFJDUPFYGLJDHNFZXJJLWMFWHKZNJGE2DCMRTGQ4TQNNHORZGSZ3HMVZKMY3SMVQXIZI> . You are receiving this email because you commented on the thread. Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub> .

just tried this, im getting "> Error: Failed to compress conversation for model switch"
…
On Mon, Oct 6, 2025 at 4:36 AM Ran Aroussi @.> wrote: @.* commented on this gist. ------------------------------ Switching to a model automatically compacts the conversation, that’s where I first encountered it … <#m_-8983328577459520007_> On Mon, Oct 6, 2025 at 4:23 AM mastertyko @.*> wrote: @.** commented on this gist. ------------------------------ Yes. What I do, is switch the model to one of FactoryAI's built-in models, run /compact, and switch back to a the custom model. I see, thank you. I will try that out. — Reply to this email directly, view it on GitHub https://gist.github.com/chandika/c4b64c5b8f5e29f6112021d46c159fdd#gistcomment-5788921 or unsubscribe https://github.com/notifications/unsubscribe-auth/BO2MNNWC2JVI5EJQWTIPLFD3WIRINBFHORZGSZ3HMVZKMY3SMVQXIZNMON2WE2TFMN2F65DZOBS2WR3JON2EG33NNVSW45FGORXXA2LDOOIYFJDUPFYGLJDHNFZXJJLWMFWHKZNJGE2DCMRTGQ4TQNNKMF2HI4TJMJ2XIZLTSOBKK5TBNR2WLKBRGEZTCMJUG442I3TBNVS2QYLDORXXEX3JMSBKK5TBNR2WLJDUOJ2WLJDOMFWWLO3UNBZGKYLEL5YGC4TUNFRWS4DBNZ2F6YLDORUXM2LUPGBKK5TBNR2WLJDHNFZXJJDOMFWWLK3UNBZGKYLEL52HS4DF . You are receiving this email because you commented on the thread. Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub . Nice find!! — Reply to this email directly, view it on GitHub https://gist.github.com/chandika/c4b64c5b8f5e29f6112021d46c159fdd#gistcomment-5788935 or unsubscribe https://github.com/notifications/unsubscribe-auth/BO2MNNTDQRGTQBT4L46LUNL3WISYHBFKMF2HI4TJMJ2XIZLTSOBKK5TBNR2WLJZRGE4DKNBVHCSG4YLNMWUGCY3UN5ZF62LEQKSXMYLMOVS2I5DSOVS2I3TBNVS3W5DIOJSWCZC7OBQXE5DJMNUXAYLOORPWCY3UNF3GS5DZQKSXMYLMOVS2IZ3JON2KI3TBNVS2W5DIOJSWCZC7OR4XAZNMON2WE2TFMN2F65DZOBS2WR3JON2EG33NNVSW45FGORXXA2LDOOIYFJDUPFYGLJDHNFZXJJLWMFWHKZNJGE2DCMRTGQ4TQNNHORZGSZ3HMVZKMY3SMVQXIZI . You are receiving this email because you commented on the thread. Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub .

tried this again, it appears switching to a factory model version and then swapping back does work!

Would that eat into the credits you get allotted during sign up to compact? I’d assume that when u run out of those, you might have the same problem again?

I assume so, but the free 40M tokens will get you a looooooooong way if you're only using them for compressing.

You underestimate how wasteful I can be with vibe coding hahah. But yeah you’re right. Especially if you use one of the cheaper models.

I assume so, but the free 40M tokens will get you a looooooooong way if you're only using them for compressing.

i always have error when switch model to do compression