Eko
Type: full-code · Vendor: FellouAI · Language: TypeScript, JavaScript · License: MIT · Status: active · Status in practice: emerging
Build production-ready agentic workflows in JavaScript/TypeScript from natural-language commands — decomposing a request into a multi-step workflow that runs multi-agent across browser and computer environments, with native MCP, human-in-the-loop, dependency-aware parallel execution, and pause/resume/interrupt with workflow recovery.
Description. Eko ('Eko Keeps Operating', FellouAI, MIT) is a TypeScript/JavaScript framework that turns a natural-language command into a multi-step agentic workflow. It runs in both browser and Node.js, ships agents for computer and browser environments, and supports multi-agent collaboration in a single task. Native MCP connects to external MCP servers. Eko 3.0 added dependency-aware parallel execution and pause/resume/interrupt with workflow recovery; 4.0 introduced chat conversations and refined agent logic. 'Stream Planning' renders the plan as it is generated; 'Loop & Listener Tasks' automate repetition. Native A2A and an Observable Chain are listed as coming soon.
Agent loop shape. Plan-then-execute workflow. From a natural-language command Eko streams a plan (Stream Planning) that decomposes the task into steps with declared dependencies; agents (e.g. BrowserAgent, FileAgent) execute the steps with awareness of dependencies for parallelism. Hooks intercept tool inputs/outputs and surface decisions for human-in-the-loop intervention. Pause / resume / interrupt controls allow workflow recovery from a snapshot. MCP servers are accessed natively; Native A2A and Observable Chain are roadmap features.
Primary use cases
- natural-language to multi-step workflow generation
- multi-agent browser and computer automation in JS/TS
- MCP-integrated workflows
- pause / resume / interrupt long-running automations with recovery
- dependency-aware parallel agent execution
Key concepts
- Workflow generation (docs) — One sentence to multi-step workflow with task-parallel execution.
- Multi-agent — Multiple agents (BrowserAgent, FileAgent, ...) in one task.
- Native MCP → mcp — Connects seamlessly with Awesome MCP Servers.
- Human-in-the-loop — Intervene when it matters most; hook system supports manual or AI intervention.
- Stream Planning → plan-and-execute — Dynamic rendering of the plan as it is generated.
- Loop & Listener Tasks — Automate repetitive tasks via loop and listener constructs.
- Pause / resume / interrupt + recovery → agent-resumption — Eko 3.0 controls for pausing, resuming, and interrupting workflows with state recovery.
Patterns this full-code implements —
- ★Browser Agent
Browser environment is one of two primary execution surfaces; e.g. stock-analysis-via-browser-automation example.
- ★Computer Use
Computer environment (Node.js + GUI actions like clicking, screenshots, form auto-fill, command execution) is supported alongside browser.
- ★★Plan-and-Execute
Stream Planning decomposes one natural-language sentence into a multi-step workflow; Eko 3.0 documents 'dependency-aware parallel execution'.
- ★★Model Context Protocol
Native MCP — connects with MCP servers as a top-level feature.
- ★★Human-in-the-Loop
Top-level feature: 'Intervene when it matters most'; hook system enables manual or AI intervention.
- ★★Agent Resumption
Eko 3.0 added pause / resume / interrupt controls with workflow recovery.
- ★★Parallelization
Dependency-aware parallel agent execution introduced in Eko 3.0.
- ★★Tool Use
8+ built-in tools (GUI click, web search, form auto-fill, command execution, screenshots); tool hooks intercept inputs/outputs.
- ★★ReAct
Inner agent step is tool-calling; framework framing emphasises plan-and-execute over ReAct. No 'ReAct' label found in the surveyed README or product page.
- ★★Scheduled Agent
Loop & Listener Tasks automate repetition; event/loop-driven rather than cron-style scheduled execution.
References
Provenance
- Last analyzed:
- Last updated:
- Verification status: needs-verification