Skip to content

getting-started.md

Latest commit

 

History

History
155 lines (114 loc) · 3.65 KB

getting-started.md

File metadata and controls

155 lines (114 loc) · 3.65 KB
Raw

Getting Started

English · 한국어

Requirements

  • Python 3.9 or newer
  • A local repository or project directory
  • One provider: Codex CLI, OpenRouter, OrcaRouter, OpenAI, Anthropic, Gemini, or an OpenAI-compatible local/hosted endpoint

Install

git clone https://github.com/LEE-Kyungjae/Palamedes.git
cd Palamedes
python3 -m pip install -e .
palamedes init

Palamedes stores project state in .palamedes/. Commit that directory when you want planning lineage to travel with the repository.

Start with Codex authentication

codex login
palamedes chat --provider codex

Start with OpenRouter

export OPENROUTER_API_KEY="..."
palamedes chat --provider openrouter

Set PALAMEDES_OPENROUTER_MODEL to override the default model.

Start with OrcaRouter

export ORCAROUTER_API_KEY="..."
palamedes chat --provider orcarouter --model orcarouter/auto

Palamedes sends streaming OpenAI Chat Completions requests to https://api.orcarouter.ai/v1. Set PALAMEDES_ORCAROUTER_MODEL or PALAMEDES_ORCAROUTER_BASE_URL to override the defaults.

Start with the OpenAI Responses API

export OPENAI_API_KEY="..."
palamedes chat --provider openai

Set PALAMEDES_OPENAI_MODEL to override the default model.

Start with Anthropic or Gemini

export ANTHROPIC_API_KEY="..."
palamedes chat --provider anthropic --model claude-sonnet-4-5

export GEMINI_API_KEY="..."
palamedes chat --provider gemini --model gemini-2.5-pro

claude is accepted as an alias for anthropic.

Start with vLLM, Ollama, LM Studio, or another compatible endpoint

Palamedes accepts any streaming OpenAI Chat Completions endpoint. Local endpoints do not require a key by default.

palamedes chat \
  --provider vllm \
  --model Qwen/Qwen3-32B \
  --provider-base-url http://127.0.0.1:8000/v1

For a protected or hosted compatible endpoint, keep the secret in an environment variable and pass only its name:

export MY_LLM_KEY="..."
palamedes chat \
  --provider openai-compatible \
  --model my-model \
  --provider-base-url https://llm.example.com/v1 \
  --provider-api-key-env MY_LLM_KEY

vllm, ollama, and lmstudio are aliases for openai-compatible; the endpoint remains explicit so Palamedes never guesses which local service owns a port. The same provider flags are available to palamedes watch.

First session

Inside the chat terminal:

/think What product assumption is most likely to be wrong?
/opportunity Find retention, revenue, distribution, and operating opportunities.
/cycle Choose the smallest reversible mission that tests user value.
/approve

Palamedes proposes mission state but does not execute implementation work.

Basic plan-state CLI

palamedes plan \
  --goal "Choose an evidence-backed direction" \
  --success-metric "Meet the preregistered metric within four weeks" \
  --planning-horizon "4 weeks" \
  --review-cadence "weekly"

palamedes evidence \
  --claim "The problem repeated in user interviews" \
  --source "interviews" \
  --confidence 72 \
  --axis market

palamedes hypothesis \
  --statement "The problem changes purchase behavior" \
  --confidence 55

palamedes show
palamedes qa
palamedes history

Multiple workspaces

palamedes workspace add my-product /path/to/project
palamedes -w my-product show
palamedes -w /path/to/project chat --provider codex

Verify the checkout

make check

For commands and recovery behavior, continue to operations reference. For reasoning modes, see cognition workflows.