Issue with Paperclip DeepSeek Integration

[HubRambo]

I spent several sessions trying to get codex_local routing to DeepSeek via their OpenAI-compatible API, and hit two separate walls that I wanted to document here in case it saves someone else time.

Problem 1 — codex_local ignores OPENAI_BASE_URL entirely
Setting OPENAI_BASE_URL=https://api.deepseek.com in adapterConfig.env does nothing — confirmed by the 401 Unauthorized errors in run logs, which prove requests are still going to api.openai.com. I also tried setting it at process level via inline env vars when starting the server (OPENAI_BASE_URL=https://api.deepseek.com npx paperclipai run). Same result, still $0 on DeepSeek, still 401s on OpenAI. It seems codex_local hardcodes or ignores the base URL entirely.

Problem 2 — opencode_local registers and discovers models but never spawns processes
Switched to opencode_local as the next attempt (OpenCode installed at /opt/homebrew/bin/opencode, DeepSeek auth configured in ~/.local/share/opencode/auth.json, model set to deepseek/deepseek-v4-flash). The adapter correctly lists DeepSeek models at /api/companies/:id/adapters/opencode_local/models. Runs enter running status. But processPid stays null, processStartedAt stays null, zero events, and the run just hangs indefinitely with no output.

Environment: macOS, Paperclip via npx paperclipai run, OpenCode v1.14.x+

Would love to know if there's a required command path or additional adapterConfig field needed to actually get opencode_local to spawn. Happy to test a fix if anyone has a lead.

[thierryteisseire]

Did setup claude code running deepseek seems to work - still testing but running.

[qanuj]

Simply ask your CEO to create an Agent with Claude Local using API key and API url and change the env in configuration to have correct api key and endpoint https://api-docs.deepseek.com/quick_start/pricing/