Cloneable control planes and sandboxes that let power users own the agent loop

Airbnb

Benchmark-agnostic framework that evolves system prompts and tool-call middleware via eval feedback using genetic or linear optimizers.

Use caseUse it to automatically lift pass rates on coding or tool-use benchmarks by editing only the harness surfaces.

AtomicBot-ai

Local-first AI agent that runs open-weight models via llama.cpp, drives browser and shell with approvals, and persists memory across sessions.

Use caseUse it to keep private multi-step agent work entirely on-device with local models and MCP tools.

bastani-inc

Verifiable coding-agent runtime that encodes processes as stages, checks, gates and approval graphs instead of free-form prompts.

Use caseUse it to define durable engineering loops with explicit checkpoints and subagent delegation for long-horizon tasks.

desplega-ai

Open-source agentic OS that lets a lead agent dispatch goals to isolated worker containers running various coding harnesses with shared memory and review gates.

Use caseUse it to run Slack- or GitHub-triggered multi-agent fleets with Docker isolation and schema-validated results.

Vercel

Experimental HarnessAgent plus harness specification with network session sandbox interface for bridge-backed coding agents.

Use caseUse it to wire Claude Code style harnesses into Vercel sandboxes with streaming and tool execution.

nightona-co

Community fork continuing the last open AGPL Daytona release as self-hostable elastic sandbox infrastructure for AI code execution.

Use caseUse it to keep a fully open sandbox runtime alive after the original project went private.

Arhan-w

Single-binary terminal AI agent with 180+ providers, automatic failover mid-session, prompt caching, parallel subagents and rewind.

Use caseUse it to keep coding sessions alive across quota limits with live provider switching and cost tracking.

Salesforce AI Research

Framework to evaluate and evolve agent harnesses at scale with DarwinX as the core evolution algorithm and verifier-backed tasks.

Use caseUse it to search for harness changes that improve capability while guarding against regressions.

Eigenwise

Lightweight modular framework for building agentic pipelines from single-purpose reusable components with clear schemas.

Use caseUse it to compose predictable LEGO-style agents and tools without heavy framework lock-in.

douglascorrea

Plugin that adds session titles, settle triage, workspace rollups and resumable chat history for agents running inside Herdr.

Use caseUse it to treat agent panes like an inbox you can settle and reopen with native session resume.

Zelmoghazy

Minimal Python example that turns a project folder into an isolated Podman sandbox for tool-calling coding agents.

Use caseUse it to prototype a secure agent loop with network-disabled containers and resource caps.