tesslate-agent

Standalone autonomous agent with 30+ built-in tools for code editing, shell execution, navigation, git, memory, structured planning, and subagent delegation. Single-process, batteries-included, zero runtime dependencies on the OpenSail orchestrator.

Install

uv tool install git+https://github.com/Tesslate-AI/tesslate-agent

Or from PyPI once published:

uv tool install tesslate-agent

Usage

# Run a task against the current directory
tesslate-agent run \
  --task "add a README.md with a one-paragraph project description" \
  --model openai/gpt-4o-mini \
  --workdir . \
  --output ./trajectory.json

The agent talks to any LLM reachable via LiteLLM. Export the provider API key for whichever model you pick:

Provider	Env var
OpenAI	OPENAI_API_KEY
Anthropic	ANTHROPIC_API_KEY
Google	GEMINI_API_KEY
OpenRouter	OPENROUTER_API_KEY
Groq	GROQ_API_KEY
DeepSeek	DEEPSEEK_API_KEY
Together	TOGETHER_API_KEY
Fireworks	FIREWORKS_API_KEY
Mistral	MISTRAL_API_KEY
Cohere	COHERE_API_KEY

LiteLLM proxy

If you run a LiteLLM proxy, point the agent at it instead of exporting a per-provider key:

export LITELLM_API_BASE="https://litellm.example.com"
export LITELLM_MASTER_KEY="sk-..."
tesslate-agent run --task "..." --model gpt-4o-mini

Project root

The agent reads/writes files under the directory given by --workdir (or the PROJECT_ROOT env var). Paths are resolved with strict containment checks — the agent cannot escape the project root via symlinks or ...

License

Apache-2.0. Copyright 2026 Tesslate AI.