NOTE: This repository is currently in alpha development, and we expect to ship further stability updates over the coming weeks

SiliSocS

A configurable, extensible framework for multi-agent social simulation and experimentation.

SiliSocS (Silicon Society Sandbox) is easy to use and is structured around the EASE decomposition (Environment, Agents, Simulation engine, and Evaluation), taking inspiration from the Concordia framework and providing a principled, reproducible configuration layer for simulated worlds. You define each axis in YAML: the environment agents inhabit, the agent population and their memories, the simulation engine that schedules how and when agents act, and the evaluation that measures them. Built-in environments span social platforms (a Twitter-like app, a Reddit-like app, and a real Mastodon server), a resource market, and a virtual space, and you can add your own. SiliSocS offers scenario-driven grounding, game master-mediated environments, local and served backends, evaluation probes, runtime telemetry, and experimental study tooling. Interoperability for Concordia-designed agents is available through an optional bridge extra.

• 2026 ICML Position Paper ICML 2026
• 2026 EASE Configuration: arXiv:2605.30258
• Documentation: sandbox-social.github.io/silisocs

Papers on a previous version centered on a served Mastodon social media network:

• 2024 NeurIPS Workshop Paper: arXiv:2410.13915
• 2025 IJCAI Demo Paper: IJCAI 2025

Install

The default package is intentionally lean and supports local simulations without Studio, Mastodon, HuggingFace, or analysis dependencies:

pip install silisocs

Optional integrations are exposed as extras:

pip install "silisocs[hf]"        # Hugging Face persona sources
pip install "silisocs[mastodon]"  # real Mastodon backend
pip install "silisocs[studio]"    # unified visual workspace
pip install "silisocs[analysis]"  # notebooks and extended analysis
pip install "silisocs[concordia]" # optional Concordia bridge
pip install "silisocs[hpc]"       # optional Submitit/Slurm study helpers

For contributor work from a checkout:

git clone https://github.com/sandbox-social/silisocs.git
cd silisocs
uv sync --all-extras --group dev --group docs

Quick Start

Run the built-in package default scenario:

uv run silisocs

For a local smoke test without model API calls:

uv run silisocs sim.llm.provider=scripted

Override scale or model settings with Hydra dot notation:

uv run silisocs num_agents=10 num_steps=5 sim.llm.name=gpt-4o

Run a bundled external scenario:

uv run silisocs --config-path scenarios/election/conf

Run the packaged resource-market preset:

uv run silisocs world=resource_market agents=resource_market env=resource_market

Run the packaged virtual-space preset:

uv run silisocs world=virtual_space agents=virtual_space env=virtual_space

The same backends also have curated external examples under scenarios/:

uv run silisocs --config-path scenarios/resource_market/conf world=resource_market agents=resource_market env=resource_market
uv run silisocs --config-path scenarios/virtual_space/conf world=virtual_space agents=virtual_space env=virtual_space

Outputs are written under outputs/<scenario_name>/<jobname>/ and include action_events.jsonl, probe_events.jsonl, prompts_and_responses.jsonl, sim_metrics.json, a resolved Hydra config snapshot, and a local SQLite backend database for local platforms.

Studies and Experiments

Study orchestration ships in the package as silisocs.studies.run_study, exposed as the silisocs-study console command (equivalent: python -m silisocs.studies.run_study). It expands hypotheses, conditions, scenarios, and seeds into reproducible simulation runs, then executes the configured evaluators and writes organized artifacts under the study's generated/ directory.

uv run silisocs-study --study experiments/studies/study_template_v1 plan
uv run silisocs-study --study experiments/studies/study_template_v1 run --only-hypothesis h1_timeline_mechanism
uv run silisocs-study --study experiments/studies/study_template_v1 summary-append --author analyst --hypothesis h1_timeline_mechanism --note "Initial finding"

Custom commands plug in through conditions.<id>.execution.command, evaluator commands through the evaluations list, and optional HPC setup through the study runner's submitit or slurm-array commands. See docs/experiments.md and docs/study_schema.md.

Architecture

The canonical runtime entry point is src/silisocs/runtime/runner.py. It composes Hydra configuration, builds agents, initializes memory, constructs the environment backend and game master, runs the simulation engine, and writes artifacts.

silisocs/
├── src/silisocs/
│   ├── agents/              # Native and bridge-compatible runtime agents
│   ├── conf/                # Packaged Hydra defaults
│   ├── studio/              # Unified visual workspace and control plane
│   ├── environments/        # Game masters and environment backends
│   ├── evaluations/         # Probes, telemetry, and optional analysis tools
│   ├── runtime/             # Runner, config projection, and orchestration
│   └── simulation_engines/  # Engine loop, step, and turn policies
├── scenarios/                  # Scenario configs and curated inputs
├── experiments/             # Study orchestration and generated study outputs
├── docs/                    # ProperDocs documentation
└── tests/                   # Unit and integration tests

Optional Concordia Bridge

SiliSocS runs on native runtime contracts by default. The optional Concordia bridge is for porting Concordia-designed agents or components without making Concordia part of the default install:

• All runtime agents satisfy silisocs.agents.base_agent.Agent.
• Native runtime classes and GM components are the primary extension API.
• Legacy Concordia-shaped components are isolated behind silisocs.adapters.concordia.
• Scenario YAML selects builders, backends, policies, probes, and prompts so most experiment designs do not require Python edits.

See docs/concordia_bridge.md and docs/building_agents.md for the extension contracts.

Development

Common commands:

uv run pytest
uv run silisocs-config-dry-run --project-root .
uv run --group dev poe lint
uv run --group dev poe docs
uv build --sdist --wheel
uv run --group docs properdocs build --strict

• uv run pytest runs the test suite in the current environment.
• uv run silisocs-config-dry-run --project-root . composes shipped scenario and replication configs without running LLM calls.
• uv run --group dev poe lint runs the configured formatting, static checks, and type checks.
• uv run --group dev poe docs runs the configured documentation build task.
• uv build --sdist --wheel builds release artifacts in dist/.
• uv run --group docs properdocs build --strict builds the documentation site and fails on broken links or stale navigation.