Generative Agents: Interactive Simulacra of Human Behavior

Agent S is an open-source computer-use agent framework from Simular. It accepts natural-language tasks, observes a real desktop through screen and accessibility signals, and operates ordinary applications by clicking, typing, scrolling, and executing optional local code rather than relying on per-application APIs or scripts. The project has progressed through three generations: accessibility-tree grounding with an LLM, a vision-based manager-worker architecture using multiple foundation models, and Agent S3, which scales computer-use attempts through parallel rollouts and selection. Its workflow can combine model-driven exploration with reusable automation code, replay saved skills, and recover when the interface changes. The framework supports macOS, Windows, and Linux, works with models from OpenAI, Anthropic, Gemini, OpenRouter, vLLM, and open-weight providers, and is distributed through the `gui-agents` Python package. An optional local coding environment can execute Python and Bash with the current user's permissions, so the repository warns that this capability should be used only in trusted environments.

RSIAgent is an open-source, training-free multi-agent framework for recursive self-improvement in unfamiliar digital environments. It uses a Curriculum Agent, Actor Agent, and Verifier Agent while keeping model parameters fixed. Its learning loop has two stages: broad recursive self-exploration, in which agents perform diverse projects and consolidate verified experience, followed by deep recursive self-exploration, in which the Curriculum Agent selects focused practice around gaps and fragile successes. The Actor Agent interacts with software through executable Python or Bash programs and visual observations; the Verifier Agent independently checks task requirements and resulting environments. Procedures, scripts, successful experiences, and failure lessons are stored in persistent memory, which is frozen and reused for downstream task execution. The repository provides runtimes and benchmark integrations for OSWorld-V2 and Agents' Last Exam, along with setup, smoke-test, batch execution, reporting, recovery, and validation utilities. It requires Python 3.12 and a Linux host with Docker and /dev/kvm for the documented benchmark runs, and is licensed under Apache License 2.0.

agentplay is an open-source command-line tool and local web viewer maintained by Entelligence AI for replaying coding-agent sessions as a map and timeline. It discovers session data from Claude Code, Codex, Cursor, GitHub Copilot, OpenCode, and Kiro, parses transcripts or local database stores into redacted agentplay/v1 JSON maps, and serves them from a local viewer. The session map records files and repositories touched, shell verbs, tools, MCP servers, models, subagents, errors, retries, delays, outputs, token counts, and timing without storing prompt or assistant text, tool inputs or outputs, file contents, or command arguments. The viewer provides a radial session map, playback with step or wall-clock time, error and retry navigation, task and subagent scopes, and import/export of map files. It runs locally without an account or API key; session data remains on the device, with an optional email-signup network request that can be disabled. The CLI requires Python 3.11 or newer, has no runtime dependencies beyond the standard library, and is distributed under the Apache License 2.0.

Agency Agents is a collection of specialized AI-agent instruction files and a native desktop app for macOS, Linux, and Windows. Its agents define domain-specific identities, workflows, rules, deliverables, code examples, and success metrics across engineering, design, marketing, sales, testing, security, finance, game development, GIS, healthcare, and other divisions. The repository provides scripts that convert and install the agents into tools including Claude Code, Cursor, GitHub Copilot, Gemini CLI, OpenCode, Aider, Windsurf, Qwen Code, Kimi Code, Codex, Osaurus, and others; the app browses the roster, installs selected agents, and keeps them updated. The repository is distributed under the MIT License.

eSlams is an open-source AI evaluation platform and Python package in which models and agents play games against one another or against people. It uses games such as chess, Go, poker, cooperation games, and arcade games to evaluate agentic action and long-chain reasoning beyond static benchmarks. The public eSlams Core engine provides dozens of games, a strict agent protocol, and deterministic, replayable matches. Its website offers Arena play against AI models, Battlefield replays of model-versus-model matches, and a Grand Slams tournament; the engine can also be installed with pip and run locally.