simulationlab 0.2.0

A self-serve simulation lab, enabling you to build and refine long-horizon AI agents.

SimLab is the data layer for adaptively composing RL simulations and evaluating and refining agents. SimLab is toolset, agent harness and sandbox agnostic. Browse pre-built scenario templates or bring your own CLI/MCP toolset.

• Browse & compose environments from a catalog of tool servers and scenario templates
• Run agents against tasks using any LLM provider (OpenAI, Fireworks, custom endpoints)
• Generate custom tasks with built-in task generation pipelines
• Evaluate automatically with verifiers and reward model scoring
• Scale to the cloud with Daytona for remote sandbox execution (if you want to experiment with large-scale parallel rollouts, reach out to us or join the Discord!)

Quickstart

Install

# Recommended: install with all extras
uv pip install "simulationlab[npc,daytona,langchain]"

# Or install only what you need:
uv tool install simulationlab
uv tool install "simulationlab[npc]"        # NPC chat simulation
uv tool install "simulationlab[daytona]"    # Remote sandbox execution
uv tool install "simulationlab[langchain]"  # LangChain/LangGraph agents

# pipx works too:
pipx install simulationlab

The PyPI package is named simulationlab. The installed CLI command is simlab. SimLab currently supports Python 3.13.

git clone https://github.com/collinear-ai/simlab.git
cd simlab/cli/simlab
uv tool install .
# or with extras:
uv tool install ".[daytona]"

uv run simlab <command>

Get your API keys and export them:

export SIMLAB_COLLINEAR_API_KEY="col_..."   # from platform.collinear.ai (Developers > API Keys)
export DAYTONA_API_KEY="dtn_..."            # from app.daytona.io
export OPENAI_API_KEY="sk-..."              # from platform.openai.com/api-keys
export SIMLAB_VERIFIER_MODEL="gpt-5.2"      # reward model (tasks also come with a programmatic verifier so can skip this)
export SIMLAB_VERIFIER_PROVIDER="openai"    # litellm compatible provider name
export SIMLAB_VERIFIER_API_KEY="$OPENAI_API_KEY" # corresponding key

Create an environment:

simlab env init my-env --template hr

To list templates run simlab templates list (or simlab templates list --env my-env to use that env's Scenario Manager URL)

Create and list tasks in directory ./generated-tasks

simlab tasks-gen init --preset recruiting # Can go to the config.toml to setup number of tasks etc.
simlab tasks list --tasks-dir ./generated-tasks # takes 5-10 mins with the default setting, choose haiku and 2 tasks for a faster generation.

Run a task with task id task_id.

tasks run automatically starts the environment, seeds data, runs the agent, and tears down when done.

simlab tasks run --env my-env \
  --task task_id \
  --tasks-dir ./generated-tasks \
  --daytona \
  --agent-model gpt-5.2 \
  --agent-api-key "$OPENAI_API_KEY"

simlab env init my-env --template hr
simlab tasks run --env my-env --task task_id --agent-model gpt-5.2 --agent-api-key "$OPENAI_API_KEY"

For the full walkthrough — configuration, custom agents, task generation, verifiers, and more — see the Quickstart Guide. For framework adapters and custom agent integration patterns, see Agent Integrations.

For environment-scoped tool definitions, see Env-Local Custom Tools.

API Keys

Configuration

SimLab resolves configuration in this order: config file < environment variables < CLI flags.

Config file: ~/.config/simlab/config.toml (override with --config-file or SIMLAB_CONFIG)

collinear_api_key = "col_..."

[agent]
model = "gpt-5-mini"
provider = "openai"
api_key = "sk-..."

[daytona]
api_key = "dtn_..."

[verifier]
model = "claude-sonnet-4-6"
api_key = "sk-ant-..."

[npc_chat]
model = "gpt-4o-mini"       # LLM model for NPC chat responses (default: gpt-4o-mini)
api_key = "sk-..."           # API key (falls back to agent key, then provider env vars)

Environment Variables

CLI Reference

Run simlab --help for full usage details.

Documentation

• Quickstart Guide — full setup and usage walkthrough
• Env-Local Custom Tools — add custom tool definitions under one environment
• Agent Integrations — adapter architecture and custom framework integration guide
• Docs — complete documentation
• Collinear Platform — get your API key

License

This project is licensed under the Apache 2.0 License.

Contact

Questions or feedback? Reach out to us:

• nazneen@collinear.ai
• sachin@collinear.ai
• adit@collinear.ai