Agent Landscape
Other people’s answers to the problems in agent-harness. Each page says what the tool does and which idea, if any, I took from it. The category map is agent-toolkit-landscape.
Running many agents
- conductor-parallel-agents: Conductor, Superset, Claude Squad and the rest: many agents in parallel worktrees
- paperclip-agent-company: agents in an org chart, with budgets that hard-stop and heartbeats in a database
- ruflo-orchestration: 32 plugins and 100+ agents in one self-learning swarm
- bmad-method: agents and workflows as Claude Code skills, persona in a TOML file
- buzz-agent-workspace: humans and agents in the same rooms, every message a signed event
Runtimes and agent servers
- cloudflare-computer-agent-runtime: a durable SQLite filesystem per agent with compute attached on demand
- deepseek-harness: every part a plugin, the loop included
- hermes-agent: an agent that writes skills from its own experience
- rust-agent-ecosystem: ZeroClaw, IronClaw, Moltis, ZeptoClaw compared
- moltis-rust-agent: a Rust agent server I contributed to
Memory and code search
- code-graph-mcp-landscape: 15 code-graph tools, nearly all ended on tree-sitter plus SQLite FTS5
- graphify-vs-solograph: tree-sitter graphs with no embeddings, against my own index
- mempalace-agent-memory: local memory scoring 96.6% on LongMemEval
- fff-agent-file-search: fuzzy file search with frecency, over MCP
- promptql-context-wiki: what makes an agent accurate on company data
- semantica-decision-provenance: a graph that answers a regulator’s “why” months later
The browser
- vercel-agent-browser: accessibility-tree references instead of selectors
- webwright-code-as-action: the agent writes Playwright scripts; the browser is disposable
Apps built for agents
- agent-native-builder: one action definition serves the UI, the agent, HTTP, MCP and CLI
- clips-agent-native-video: a Loom alternative an agent can read and edit
- agent-native-format-choice: video rendered from plain HTML, because agents already write HTML
Research agents and safety
- asi-evolve-ai-research-agents: agents running full-cycle AI research
- shinka-evolve: code evolved with LLMs as the mutation operator
- dsewiki-agent-collusion: rogue agents used a public wiki to share ways to cheat their evals