Skip to content

Using PlanExe with Anthropic

Anthropic makes the Claude family of models. PlanExe can use Anthropic models directly via the Anthropic API.

There are two ways to authenticate:

Method Who it's for Token format
API key Anthropic API billing account sk-ant-api*
Claude Code OAuth token Claude Pro/Max subscribers sk-ant-oat*

The Claude Code OAuth path lets you use PlanExe without a separate API billing account — if you already pay for Claude Pro or Max, you can generate a token and use it directly.

Authentication setup

Option A — Standard API key

  1. Sign up at console.anthropic.com and add API credits.
  2. Generate an API key from the Console.
  3. Add to your .env: 
    ANTHROPIC_API_KEY=sk-ant-api...

Option B — Claude Code OAuth token (Claude Pro/Max subscribers)

If you have a Claude Pro or Max subscription, you can generate an OAuth token that works in place of an API key.

Requirements: - Claude Code installed (npm install -g @anthropic-ai/claude-code) - Active Claude Pro or Max subscription

Steps:

  1. Run: 

    claude setup-token
    This outputs a token starting with sk-ant-oat.

  2. Add to your .env: 

    ANTHROPIC_API_KEY=sk-ant-oat01--...

PlanExe auto-detects the sk-ant-oat prefix and switches to the correct Bearer auth scheme automatically. No other changes needed.

Note: OAuth tokens may expire. If you get 401 errors, run claude setup-token again to refresh.

Choosing a model

PlanExe makes 60–100+ LLM calls per plan. Choose based on your priority:

Config ID Model Speed Cost (est. per plan) Notes
anthropic-claude-haiku-4-5 Haiku 4.5 Fastest < $0.50 Best default for most users
anthropic-claude-sonnet-4-6 Sonnet 4.6 Medium ~$2–5 Better quality than Haiku
anthropic-claude-sonnet-4-6-thinking Sonnet 4.6 + thinking Slow ~$5–15 Extended reasoning per call
anthropic-claude-opus-4-6 Opus 4.6 Slow ~$15–40 Highest quality, highest cost
anthropic-claude-opus-4-6-thinking Opus 4.6 + thinking Very slow ~$30–80+ Maximum quality, use sparingly

For pinned versions (reproducible results), use: - anthropic-claude-haiku-4-5-pinned — locked to claude-haiku-4-5-20251001 - anthropic-claude-sonnet-4-5 — locked to claude-sonnet-4-5

Quickstart (Docker)

  1. Install Docker (with Docker Compose).
  2. Clone the repo: 
    git clone https://github.com/PlanExeOrg/PlanExe.git
cd PlanExe
  3. Copy .env.docker-example to .env and configure: 
    ANTHROPIC_API_KEY=sk-ant-...
DEFAULT_LLM=anthropic-claude-haiku-4-5
PLANEXE_MODEL_PROFILE=custom
PLANEXE_LLM_CONFIG_CUSTOM_FILENAME=anthropic_claude.json
  4. Start PlanExe: 
    docker compose up worker_plan frontend_single_user
  5. Open http://localhost:7860 and submit a prompt.

To switch models, change DEFAULT_LLM in .env to any config ID from the table above, then restart: 

docker compose restart worker_plan frontend_single_user

Extended thinking

Some entries in anthropic_claude.json enable extended thinking. These use a thinking_dict in the config:

"thinking_dict": {"type": "enabled", "budget_tokens": 10000}

budget_tokens controls how many tokens Claude can use for internal reasoning before writing its answer. Higher = slower + more expensive, but better structured output on complex tasks.

Trade-offs for PlanExe: - PlanExe runs 60–100+ calls per plan — thinking multiplies both cost and latency significantly - For most plans, Haiku or Sonnet without thinking produces good results - Use thinking variants only when plan quality is the top priority

Troubleshooting

401 Unauthorized - API key: check that ANTHROPIC_API_KEY is set correctly in .env. - OAuth token: run claude setup-token again — tokens expire.

400 credit balance too low - API key path: add credits at console.anthropic.com. - OAuth token path: OAuth tokens authenticate you, but API calls still consume credits from your Anthropic account. Top up at the Console if needed.

Check worker logs: 

docker compose logs -f worker_plan

Report issues on Discord. Include your OS, hardware, and the error from log.txt in the output directory.

Next steps