One sentence. Full team. Auto delivery.
AI-Native orchestrator with persistent agent identity, shared team learning, retry/circuit-breaker resilience, atomic operations, scheduled 24/7 execution, code verification, and OpenAPI tool bridge. From any goal to a working solution.
Smart Orchestrator analyzes your goal, builds the perfect team, executes in parallel, then synthesizes results.
Agents move between zones as they plan, code, debug, and test -- all visible in a pixel-art virtual office.
Run superagent auto "goal" --command builtin --dashboard to see it live
A full pixel-art virtual office -- agents move between zones as they work.
Idle → Writing → Executing → Bug Fix → Syncing → Complete -- all in a cozy pixel office
Ingest documents into an LLM Wiki. Explore connections with an interactive vis.js graph.
superagent wiki ingest docs/ then superagent wiki graph --format html
Scale meets simplicity.
Any goal. Smart team. Parallel execution. Real results.
LLM analyzes your goal and dynamically picks the best roles from 178 templates across 14 departments.
superagent auto "Make $100" --smartMythos pattern: workers run concurrently via asyncio.gather. 90%+ speedup on multi-agent tasks.
superagent auto "Build an app" --parallelPersistent agent identity with experience tracking, domain expertise, and correction learning. Agents grow smarter over time. (Moxt pattern)
superagent auto "goal" --command builtinFix one agent, all agents learn. Corrections propagate across the entire team via shared memory injection. (Moxt pattern)
superagent corrections add "Use TypeScript"Cron-style agent task automation. Schedule nightly monitoring, weekly reports, hourly analysis. (Moxt pattern)
superagent schedule add "Monitor" --cron "0 2 * * *"Send new requirements to running teams mid-execution. Leaders re-evaluate and incorporate directives at synthesis checkpoints.
superagent inject send team "Add dark mode"Exponential backoff retry (3x), per-provider circuit breaker, atomic file writes, thread-safe SQLite, transaction batching.
Auto retry on 429/5xx + circuit breakerVerify agent-generated code locally with timeout, output capture, file detection, and guardrail safety checks. (OpenAgents pattern)
superagent exec "print(42)"Import any tool from ai-plugin.json + OpenAPI specs. ChatGPT/OpenAgents plugin standard compatible. (OpenAgents pattern)
Standard plugin manifest supportKarpathy-style knowledge base: ingest docs, extract entities, cross-reference with wiki-links, query with LLM.
superagent wiki ingest paper.pdfA real browser agent that navigates, clicks, extracts data from any website.
superagent browser run "Find trending AI"Input + output safety checks, blocks destructive commands, scans imported skills, sandboxed execution via E2B.
superagent guardrail listRecords outcomes, learns from failures, auto-evolves. Skill packs for community sharing.
superagent skill auto-evolvePixel-art office + Hermes HUD Web UI. Agents move between zones in real-time as they work.
superagent auto "goal" --dashboardPersistent HTTP server with REST API + mobile-friendly Web UI. Monitor and control teams from your phone. (CodeBuddy/OpenClaw pattern)
superagent serve --port 7890Auto-snapshots before each agent edit. Revert unwanted changes with one click. Your safety net for autonomous execution. (CodeBuddy pattern)
superagent checkpoint list --session s1Bidirectional messaging with WeCom, DingTalk, Feishu, and webhooks. DM pairing auth, inbound queue, and reply routing. (CodeBuddy/OpenClaw pattern)
POST /api/v1/channels/webhook/wecomRun tasks in background with ps/logs/kill. Daemon mode for persistent service. PID file registry with auto-cleanup. (CodeBuddy pattern)
superagent worker bg "Refactor auth"Run everything with Ollama/LocalAI/llama.cpp. No API keys, no costs.
superagent auto "goal" --command localMulti-platform content publishing: Medium, Dev.to, WordPress, Zhihu, Juejin.
superagent publish article.mdInteractive AI coding assistant with streaming responses, tool execution, slash commands, and session persistence. Like CodeBuddy Code in your terminal.
superagent chatProject-level instruction cascade. Like CLAUDE.md and CODEBUDDY.md — deeper files take priority, section-aware merge across directories.
SUPERAGENT.mdShared endpoint presets and reusable agent runtime profiles. Configure once, deploy across teams and projects.
superagent preset list / superagent profile list13 built-in LLM provider presets with auto-discovery. LLM Gateway picks the best model per task based on category and complexity.
superagent llm list / superagent auto "goal"Create pull requests from uncommitted changes with one command. Auto-report persistent failures as GitHub Issues.
superagent pr "Fix auth" / superagent issue reportStanding on the shoulders of 577k+ stars.
From any goal to delivery in one pipeline.
superagent auto "goal" --smart --command builtin --dashboard
|
+-- Smart Orchestrator (LLM-powered goal analysis)
| +-- Complexity detection (simple/moderate/complex)
| +-- Dynamic role matching from 178 templates
| +-- Team composition with permission levels
|
+-- Execution Engine (3-phase Mythos pattern)
| +-- Phase 1: Leader planning
| +-- Phase 2: Workers parallel execution (asyncio.gather)
| +-- Phase 3: Synthesis review (quality + completeness)
|
+-- Knowledge Layer
| +-- LLM Wiki (Karpathy: raw -> wiki -> index)
| +-- 3-layer memory (conversation/persistent/skill)
| +-- Interactive knowledge graph (vis.js)
| +-- SHA256 incremental cache
|
+-- Safety Layer
| +-- Guardrails (block rm -rf, drop table)
| +-- E2B Sandbox (isolated execution)
| +-- Skill threat scanning
|
+-- Visualization
+-- Dashboard (pixel-art virtual office)
+-- Knowledge graph (HTML interactive)
+-- Rich terminal outputFrom zero to autonomous agents in 3 steps.
pip install superagent-clisuperagent auto "Build a REST API" --smart --command builtin# Open dashboard:
superagent auto "goal" --dashboard
# Build knowledge:
superagent wiki ingest docs/
superagent wiki graph --format html