A Claude Code alternative that runs on the model you pick
Keep the terminal loop you know: read the repo, edit files, run the tests, report back. Run it on DeepSeek, GLM, GPT, Claude or a model on your own machine, from one open-source binary.
curl -fsSL https://octomind.run/install.sh | bash Verdict: stay with Claude Code or switch to octomind
Stay with Claude Code if
- You review every edit in a side-by-side diff in VS Code or JetBrains, or undo a turn with /rewind. Octomind has neither.
- Your Claude Pro or Max plan already covers your usage. Claude Code comes with every paid Claude plan, and per-token bills would be new for you.
- You only want Anthropic models, run by the harness Anthropic builds for them.
- You want the agent to ask before it edits a file or runs a command. Octomind has no approval prompts.
Switch to octomind if
- You want the same workflow on non-Claude models: DeepSeek, GLM, GPT or a local model through Ollama or LM Studio. Anthropic does not support routing Claude Code to non-Claude models.
- You need source you can read, fork and ship inside your own image. Octomind is Apache-2.0; Claude Code is proprietary.
- You want a per-request spend stop that also works in interactive sessions, and a distinct exit code (75) when a CI run hits it.
- You are on the Claude Free plan, which does not include Claude Code. The octomind binary is free, and a local model adds no API bill.
Claude Code vs octomind
| Aspect | Claude Code | octomind |
|---|---|---|
| Price to start | Not on the Free plan. Pro is $20 a month billed monthly ($17 a month billed yearly), Max starts at $100 a month, or pay per token with a Claude Console account. Prices as of October 2026. claude.com | The binary is free under Apache-2.0. You pay your model provider per token, or nothing in API fees with a local model. Cargo.toml:6 |
| Models you can use | Claude models only, through Anthropic, Amazon Bedrock, Google Cloud or Microsoft Foundry. Anthropic does not support routing Claude Code to non-Claude models through a gateway. code.claude.com | Any provider:model string with -m or /model: anthropic:, openai:, deepseek:, openrouter:, local: for Ollama or LM Studio, and more. You can switch mid-session. src/commands/run.rs:46 |
| Source code | Public repository, but proprietary: "All rights reserved", use subject to Anthropic's Commercial Terms. github.com | Apache-2.0. Read it, fork it, build it with cargo. Cargo.toml:6 |
| Project instructions | CLAUDE.md. Since v2.1.277 it also reads AGENTS.md when there is no CLAUDE.md in the working directory or above it. code.claude.com | AGENTS.md in the directory you start in. CLAUDE.md is not read, so rename it and both tools share one file. src/session/chat/session/utils.rs:102 |
| MCP servers | stdio, HTTP, SSE (deprecated) and WebSocket, configured in .mcp.json or ~/.claude.json. code.claude.com | stdio and Streamable HTTP, with OAuth discovery, configured as [[mcp.servers]] in TOML. No SSE. /docs/usage/07-mcp-tools |
| Spend cap | --max-budget-usd stops a run once estimated spend reaches the amount, in print mode (-p) only. Spend can pass the cap. code.claude.com | max_request_spending_threshold stops a request in interactive and headless runs. A headless run that hits it exits with code 75. An in-flight call can still overshoot. src/session/chat/session/messages.rs:162 |
| Reviewing and undoing edits | Side-by-side diffs with per-change accept or reject in VS Code. /rewind restores code, the conversation or both to an earlier prompt. code.claude.com | A plain line prompt with markdown output. Edits land on disk directly. There is no diff view and no rewind, so git is your undo. config-templates/default.toml:78 |
| Editors | VS Code extension and a JetBrains plugin that opens changes in the IDE diff viewer. code.claude.com | Any Agent Client Protocol (ACP) client through octomind acp. The docs show a Zed config. No VS Code extension. /docs/usage/12-editor-integration |
Claude Code on claude-opus-5: 23/25. Octomind on glm-5.2: 24/25.
On octobench, 25 tasks taken from merged pull requests in open-source projects and checked by hidden tests, each agent ran its stock setup. These are different agent-and-model pairs, not the same model in two harnesses. Claude Code ran claude-opus-5 and solved 23 of 25 for $81.79 in 6.7h. Octomind ran the open model glm-5.2 and solved 24 for $63.43 in 3.6h. Read the costs with care. About 97% of Opus context traffic was billed at Anthropic's cache-read price. The GLM runs went through Ollama cloud, which has no prompt caching, so their dollar figures are a worst case. What the run shows is that an open model in octomind landed in the same range as Claude Code on Opus. It does not show that GLM matches Opus.
Full octobench results and how to reproduce them →How to switch from Claude Code
Install octomind and pick a model
One script installs the binary to ~/.local/bin on macOS or Linux; Homebrew and cargo work too. Export a key for any provider you use. For a model on your own machine, use local: and no key.
curl -fsSL https://octomind.run/install.sh | bash export DEEPSEEK_API_KEY=sk-... # or ANTHROPIC_API_KEY, OPENAI_API_KEY, OPENROUTER_API_KEY # local model: Ollama on localhost:11434, nothing to export # octomind run developer:general -m local:qwen3-coderBring your AGENTS.md and MCP servers over
Run git mv CLAUDE.md AGENTS.md. Claude Code v2.1.277+ reads AGENTS.md when no CLAUDE.md exists, so both tools share one file. Copy each server from .mcp.json into a TOML file in the octomind config directory, and auto_bind it to the agent you run. Hooks, commands and subagents under .claude/ do not carry over.
# ~/.local/share/octomind/config/mcp-playwright.toml # was in .mcp.json: "playwright": { "command": "npx", "args": ["@playwright/mcp@latest"] } [[mcp.servers]] name = "playwright" type = "stdio" command = "npx" args = ["@playwright/mcp@latest"] timeout_seconds = 30 tools = [] auto_bind = ["developer:general"]Run your first job, with a spend stop
Run octomind run developer:general in the repo for an interactive session. For scripts and CI, pipe the task in. The -c flag caps each request in US dollars, and a run that hits the cap exits with 75 so your script can tell a budget stop from a failure.
cd ~/your/project octomind run developer:general -m deepseek:deepseek-flash # interactive echo "The test suite fails. Find the bug, fix it, run the tests." \ | octomind -c max_request_spending_threshold=0.50 \ run developer:general -m deepseek:deepseek-flash --format plain echo "exit $?" # 75 = spend stop
Other options worth a look
opencode
An MIT-licensed terminal coding agent that connects 75+ providers through Models.dev, local models included. It suits you if you want open source and a wide provider list.
Codex CLI
OpenAI's Apache-2.0 terminal agent. You sign in with a ChatGPT plan or use an API key. It suits you if you already pay for ChatGPT and want GPT models.
Aider
A pair programmer in the terminal that works with almost any LLM, local ones included, and commits each change to git with a message. It suits you if you want every edit as a commit.
Where Claude Code is better
No diff review and no rewind
Octomind writes edits straight to disk from a plain line prompt. Claude Code shows diffs in the IDE and restores earlier states with /rewind. Commit before a big task and review with git diff. Read more →
No approval prompts before tools run
Tools run without asking. Policy lives in .agents/guardrails.toml as guards that block matching calls, and --sandbox limits writes on Linux and macOS. The sandbox is off by default and does nothing on Windows. Read more →
Editor support is ACP only
There is no VS Code extension or JetBrains plugin. Octomind speaks the Agent Client Protocol, and the docs show a Zed setup. In ACP mode it sends no permission requests and writes files itself, not through the editor. Read more →
-m changes only the main model, and some defaults are hosted
Supervisor and compression calls stay on octohub:auto until you set [supervisor.model] and [compression.model]. Telemetry is on by default. The first run of developer:general clones the agent registry and installs its tool binaries, so it needs the network. Read more →
Your .claude/ setup does not come with you
Octomind reads AGENTS.md only from the directory you start in, not CLAUDE.md or parent folders. Claude Code hooks, slash commands, subagents and skills need rebuilding as guardrails, tap agents or .agents/skills/. Read more →
Questions
What is a good Claude Code alternative that works with any model?
Octomind runs the same read, edit, test loop in your terminal on any provider:model string, including DeepSeek, GLM, GPT, Claude through the API, and local models. opencode and Aider also work with many models. Claude Code itself supports Claude models only.
Is there an open source Claude Code alternative?
Yes. Octomind and Codex CLI are Apache-2.0, opencode is MIT, and Aider is open source too. Claude Code has a public repository, but its license reserves all rights to Anthropic.
Is there a free Claude Code alternative?
Claude Code is not part of the Claude Free plan. The octomind binary is free. With a local: model through Ollama or LM Studio you pay no API fees, only for your hardware. In that case set the supervisor and compression models to local too, and know that the cost line still shows a cloud-equivalent reference price.
Can I still use Claude models in octomind?
Yes. Use -m anthropic:<model> with an ANTHROPIC_API_KEY, which bills per token through the Anthropic API, or go through OpenRouter.
Do my CLAUDE.md and MCP servers carry over?
Rename CLAUDE.md to AGENTS.md; octomind reads it, and Claude Code v2.1.277 and later reads it when no CLAUDE.md exists. Each MCP server moves from .mcp.json into a [[mcp.servers]] TOML entry with the same command and args, or the same URL. Hooks, commands and subagents under .claude/ do not carry over.
How does spend control compare to --max-budget-usd?
Claude Code's --max-budget-usd works only in print mode and can be passed by in-flight spend. Octomind's max_request_spending_threshold stops a request in interactive and headless runs, and a headless run that hits it exits with code 75. It can also overshoot by one in-flight call.
More comparisons
- OpenClaw alternative A personal agent you self-host as one binary, or hand to octos on Cloud
- opencode alternative Same open model as opencode: 24 of 25 tasks solved to its 19, at about half the cost.
- Hermes Agent alternative Lessons only from your corrections, as plain files you can edit, on your box or on Cloud
- Claude Cowork alternative Hand off the job once or daily, on any model; the result lands in Telegram, Slack or email
- Manus alternative Sourced reports and live pages from a cloud machine, on one balance that stops
- Codex alternative Keep your ChatGPT plan, add any other model, and stop a request at a dollar limit.
- Cursor alternative The agent without the editor: any model, your own keys, open source, next to any IDE.
- GitHub Copilot alternative The open-source agent for terminal and CI, on any model, billed by your own provider.
- Antigravity alternative An open-source terminal agent for Gemini, Claude or local models, no Google sign-in.
- DeepSeek Harness alternative DeepSeek models in a terminal and CI agent with tagged releases and a spend stop.
Reading
Try it on the job Claude Code does for you now.
Apache-2.0 and any model. Install it, then follow the three steps above.
curl -fsSL https://octomind.run/install.sh | bash