MorseSign inOpen Morse

Use Claude Code and Codex with open models, privately

The Morse teamPublished Facts checked 6 min read

You can run Claude Code and Codex on open-weight models, privately, by pointing both at Morse Private Inference, a private API whose one address speaks Anthropic's Messages format and OpenAI's Responses format. Every request then goes to a model such as Kimi K3.1 Pro or GLM 5.3, at the privacy tier you pick, and Morse keeps no prompt and no reply.

Both tools are built to take another endpoint. Anthropic's docs describe ANTHROPIC_BASE_URL as the setting that routes Claude Code's requests "through a proxy or gateway," and Codex's config reference lets you define your own model provider with a base_url and an env_key. We read both companies' docs, plans, and data pages on November 20, 2026, and every claim about them below is quoted from those pages, listed under Sources at the end.

Morse, which publishes this page, makes a private AI chat app with GPT, Claude, Gemini, DeepSeek, and uncensored models, encrypted so even Morse can't read your chats. For developers it runs Morse Private Inference: pay-as-you-go access to 80+ open-weight models, with no subscription, and your first top-up gets 50% extra. Besides OpenAI's Chat Completions, it answers Anthropic's POST /v1/messages and OpenAI's POST /v1/responses, the two formats Claude Code and Codex speak. Every request runs at the privacy tier you choose, from a provider under a zero data retention contract to hardware no one else can read, and Morse calls each provider from an attested enclave whose code is public.

Why run a coding agent on Morse?

A coding agent sends your code out on every turn: the files it reads, the diffs it writes, and the output of the commands it runs. Running it on Morse changes three things about where that code goes.

  • It stays private. Morse keeps no prompt and no reply on any tier, never tells a provider who you are, and can hold every request to a privacy floor you set on the key.
  • It runs on open models. Kimi K3.1 Pro, Kimi K2.7 Code, GLM 5.3, GLM 5.3 Flash, and DeepSeek V4 Pro all take tools and read a lot of code at once, and you switch between them by changing one model id.
  • One address serves both tools. Claude Code and Codex use the same key, the same credits, and the same privacy settings, so a team can run both without a second account or a second set of rules.

Claude Code's requests go to POST /v1/messages and stream back in Anthropic's format, with tool calls, and a model that reasons streams its thinking as it goes. Codex's go to POST /v1/responses, its apply_patch edits come back as it expects, and every Response Morse returns says store: false. Anthropic's gateway page adds that while a gateway credential is set, "the subscription's usage limits don't apply" to those requests, since they're billed to the credential's owner, here your Morse credits.

How do you set them up?

Each tool takes a few lines, and Morse's docs give them step by step, with a check that the first edit went through:

Both pages list the coding models with their context and prices, and the models page has every model Morse serves.

How does paying per token compare with a Claude or ChatGPT plan?

On Morse you pay each model's listed price per token from credits, with no subscription. Claude Code and Codex also come with their makers' plans, which count usage in five-hour and weekly windows, and both makers sell API access by the token. Here is each as its maker describes it on November 20, 2026.

OptionHow you payModelsTraining on your prompts
Morse Private InferenceEach model's listed price per token, from credits, with no subscription80+ open-weight modelsMorse keeps no prompt or reply to train on
Claude Pro or Max$20 a month ($17 a month billed yearly), or Max from $100 a month for "5x Pro usage"Claude modelsConsumer plans train "when this setting is on"
Claude APIPer token: Opus 5.5 at $4 / $20 and Sonnet 5.5 at $2 / $10Claude modelsNot under commercial terms, unless the customer opts in
ChatGPT Plus or Pro$20 a month for "Expanded Codex usage," or Pro "From $100 / month"OpenAI modelsPersonal accounts follow the ChatGPT training preference
OpenAI API keyPer token, at API prices, "not your ChatGPT allowance"OpenAI modelsNot by default

Anthropic's usage article says paid plans have "a session limit that resets every five hours, and weekly limits that reset once every week," and that usage across claude.ai, Claude Code, and Claude Desktop counts toward the same limit. OpenAI's Codex pricing page says "ChatGPT Work and Codex share pricing, credits, and usage limits." Our guide to whether the OpenAI API is private covers what OpenAI keeps from API requests.

What does Morse keep from a coding session?

Morse keeps no prompt and no reply, on any tier, for any plan, so the code Claude Code or Codex reads never sits on Morse's side. To bill you, Morse records the model, the endpoint that served it, the token counts, the cost, how long the request took, when, and which key sent it.

  • No name on the request. A provider sees Morse, never your account, your card, or your address.
  • No fields that link your requests. Morse drops user, safety_identifier, and prompt_cache_key before a request leaves, so a provider can't tie one person's requests together.
  • No conversation stored. Every Response on POST /v1/responses says store: false.

Claude Code keeps its own copy on your computer: Anthropic's data usage page says it saves "session transcripts locally in plaintext under ~/.claude/projects/ for 30 days by default." Its environment variables page also lists CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC, which turns off "auto-updates, telemetry, error reporting," and the /feedback command, for anyone who wants nothing but the model requests to leave the machine.

Can every coding request run on attested hardware?

Yes. Give the key a privacy floor of tee in the console, and every request it makes, from Claude Code or Codex, runs inside a hardware enclave Morse checks before forwarding, or is refused. Morse never sends a request below the floor. Use z-ai/glm-5.3 or z-ai/glm-5.3-flash at that floor, since their TEE endpoints take tools, and an agent calls tools on almost every turn.

The floor lives on the key, so neither tool needs to send any extra field. A key with no floor runs each request on the model's endpoints under a zero data retention contract, and the privacy tiers page explains what each tier protects.

How do you check Morse before you send it code?

Run npx @askmorse/verify inference.trymorse.ai. It checks that the Morse Enclave runs on genuine AWS Nitro hardware, that its image is on Morse's signed list, and that it runs the open-source code at github.com/trymorse/morse-enclave. Verify it yourself walks through each step, and what is a trusted execution environment explains the hardware underneath.

The same enclave carries every message in the Morse chat app, where your history stays on your device, encrypted. The Morse docs cover streaming, tool calls, and errors, or make a free account and create your first key.

Morse publishes this guide. We've tried to keep it fair; if something's wrong or out of date, email hello@trymorse.ai.

Claude Code is a trademark of Anthropic, PBC. Codex is a trademark of OpenAI. Morse is not affiliated with or endorsed by Anthropic or OpenAI.

Sources

All read November 20, 2026.

Common questions

Can Claude Code use models other than Claude?

Yes, through any API that speaks Anthropic's Messages format. Claude Code sends its requests to the address ANTHROPIC_BASE_URL names, and Morse answers them on the open-weight model you choose.

Does Anthropic support Claude Code on non-Claude models?

Anthropic's gateway docs say any gateway that exposes a supported API format works, and that Anthropic doesn't support routing Claude Code to non-Claude models through any gateway. Questions about the model and the API go to the API's maker, which for Morse is hello@trymorse.ai.

Can Codex use a custom model provider?

Yes. Codex's config reference lets you define a provider under model_providers with its own base_url and env_key, and its models page says a custom provider needs a compatible Responses API endpoint. Morse serves POST /v1/responses.

Do I need a Claude or ChatGPT subscription to do this?

No. With a Morse key, requests are paid from your Morse credits at each model's listed price. Anthropic's docs say the subscription's usage limits don't apply to traffic sent with a gateway credential.

Which open model is best for Claude Code?

Kimi K3.1 Pro is the model Morse's docs use in their examples: it reasons, reads images, and has a 1M-token context. Kimi K2.7 Code starts writing sooner because it doesn't reason, and GLM 5.3 Flash suits quick edits at $0.15 per million input tokens.

Does Morse keep my code?

No. Morse keeps no prompt and no reply on any tier. For billing it records the model, the token counts, the cost, the time, and which key sent the request, never the content.

Can Claude Code and Codex share one Morse key?

Yes. Both tools reach the same address with the same key, so they draw on the same credits and follow the same spend limit and privacy floor set on that key.

Where are the setup steps?

Morse's docs have a page for each tool: Claude Code takes three environment variables, and Codex takes one provider block in its config.toml. Each page lists the coding models and a quick check that the first edit worked.

Can I make every coding request run in a trusted execution environment?

Yes. Give the key a privacy floor of TEE in the console and use z-ai/glm-5.3 or z-ai/glm-5.3-flash, whose TEE endpoints take tools. Morse serves each request at that floor or refuses it, and never sends it lower.

How do I check what's running before I send code?

Run npx @askmorse/verify inference.trymorse.ai. It checks the Morse Enclave's AWS Nitro attestation and that it runs the open-source code published at github.com/trymorse/morse-enclave.

Ask what you actually want to ask.

80+ models, characters, and images, with every chat encrypted and kept on your device. Free to start.

Start a private chat

Read next