Tools and harnesses
Orchestrators, harnesses, swarm platforms, infrastructure, and the evals that check whether a swarm actually did better.
19 lit particles in a tight formation, one for each toolPoint at a row to find its particle
| Tool | Category | What it does | Source |
|---|---|---|---|
repo.ing |
Swarm platform | Open swarms on hard problems: a rider names the goal, puts in a few dollars of compute, and steers by what the swarm finds. | repo.ing |
frontierharness.org |
Eval | Open benchmark for coding-agent harnesses. v1.1 covers 21 harnesses × 5 open models × 62 tasks, reporting pass rate and cost per pass. | ai.jenquist.com |
antigravity.google |
Harness | Slash command (preview) that turns a conversation into a coordinator of named parallel workers with a shared message board. | antigravity.google |
github.com |
Orchestrator | Microsoft Research’s open-source decentralized coding harness: no central orchestrator, evaluated with up to 1,024 agents. | microsoft.com |
swarmchat.io |
Coordination | Open-source, Slack-like messaging server for agent swarms, with channels, DMs, threads and a live monitor for humans. | swarmchat.io |
nanocode.xyz |
SDK | SDK for embedding Codex and Claude agents with subagents, durable sessions and remote execution. Runs on Cloudflare Durable Objects. | github.com |
amphoradb.experiments.boundless.network |
Infrastructure | Experimental serverless Postgres for agent swarms. Wallet-based auth, no API keys, pay per statement. | x.com |
github.com |
Orchestrator | Python multi-agent framework with sequential, concurrent, hierarchical and graph swarm architectures. | github.com |
t3.codes |
Control plane | Open-source control plane for coding agents. Delegates tasks to child agents in isolated git worktrees. | github.com |
idagents.ai |
Orchestrator | “Swarm OS” for running Claude Code, Codex and Cursor agents together from a single terminal. | x.com |
github.com |
Simulation | Open-source swarm-intelligence engine that simulates grounded agent personas across social and prediction-market platforms. | miroshark.xyz |
grep.ai |
Swarm platform | Managed swarms behind a manager agent with budget caps, inboxes and MCP Tasks/Events for other agents to launch them. | agentrun.ai |
cursor.com |
Research harness | Cursor’s planner/worker swarm design. It built SQLite in Rust from the docs and showed how mixing models changes cost. | cursor.com |
github.com |
Eval | Anthropic’s kit for baselining a multi-agent setup and testing whether folding subagents into one loop does better. The counterweight to swarms. | claude.com |
swarms.ai |
Eval | The same task run through a single agent and four swarm architectures, with the per-run bill for each. | swarms.ai |
botnet.com |
Eval | A method for measuring swarm output as accepted outcomes divided by all tokens spent, failed attempts included. | botnet.com |
github.com |
Orchestrator | Open-source layer from HKUDS that has a host agent dispatch parallel work to 12 coding CLIs (Codex, Claude Code, Cursor, Grok Build and others), then verify and merge the results. | github.com |
liveswarm.dev |
Plugin | Free MIT Claude Code plugin that draws each agent and subagent as a live cursor on your app. Runs locally, with no accounts or telemetry. | github.com |
swarm-storm.app |
Orchestrator | MIT agent skills for Claude Code, Codex and Cursor. Parallel coding agents settle shared files before merging, and QA agents test the running app. | github.com |
Nothing matches that search. Know one we’re missing?
Missing someone or something? Submit it with a public source.