Raven is a Python host agent that generates task DAGs and delegates work to four built-in agents (Research, Code, Design, Oncall) or to third-party agents connected over ACP, CLI or OpenAI-compatible APIs. It uses EverOS for memory across sessions, and an experimental Curator rewrites agent strategy modules. A trajectory view shows multi-agent runs as a timeline.
Who it is for: Engineers orchestrating several coding and research agents in one workflow
Strengths
Presets for 13 third-party agents (Claude Code, Codex, OpenCode, Hermes and others)
Four built-in agents: research, coding, visual design, unattended long-running workflows
Trajectory view shows every agent, turn and step of a run on one timeline
Agent loop split into memory, planning, capability and action modules
Weaknesses
Curator self-improvement is experimental and ships with the repo, not the package
Benchmark claims are self-reported and shown only as images
Memory relies on the separate EverOS project
Install is a curl-pipe-to-bash script; hardware needs are not listed
AutoGPTRole-based AI agents coordinated by one lead agent, plus a visual workflow builder
Live demo ↗
(opens in a new tab)188k stars+853 (+0.5%) in 30 dayscustom license
95 out of 100score
2
daily_stock_analysisLLM-driven daily analysis of watchlist stocks across six markets, pushed to chat apps69.9k stars+3,683 (+5.6%) in 30 daysMIT
95 out of 100score
3
LangflowVisual builder for AI agents and workflows, exposed as API or MCP server156.5k stars+1,251 (+0.8%) in 30 daysMIT
94 out of 100score
#4
deer-flowAgent harness running sub-agents, memory and sandboxes with pluggable skills85.2k stars+1,752 (+2.1%) in 30 daysMIT
93 out of 100score
#5
DifyVisual builder for LLM workflows, RAG pipelines and agents, with APIs
Live demo ↗
(opens in a new tab)161k stars+3,111 (+2.0%) in 30 dayscustom license