flyto-core is the open-source execution engine of Flyto2. You describe a job as a YAML procedure made of steps that call registry-backed modules (browser.goto, browser.screenshot, browser.performance, file.write, and so on), and the engine validates the schemas, runs the steps, and records a trace of each one with its input, output, timing, and status. When a step fails, flyto replay --from-step N re-executes from that point with earlier results preserved instead of rerunning the whole job.
It targets AI-agent builders who want agents to call reviewed, schema-defined modules rather than generate ad-hoc code. The same engine is reachable from the CLI, an MCP server, an HTTP API, and Python, and it comes with 41 built-in recipes for tasks such as competitor pricing capture with screenshots and Web Vitals, full site audits for SEO, accessibility, and performance, and scraping to CSV. Every run produces evidence snapshots that can be packaged into reports.
Flyto2 is delivered as three packages: flyto-ai for routing and governing work and provider use, flyto-blueprint for storing and scoring reusable procedures, and flyto-core for validation, execution, replay, and evidence. flyto-core runs standalone without the other two. Extensions come in two kinds, module packs (flyto-modules-) and plugins (flyto-plugin-), admitted by name prefix and entry-point group with installation off by default.
Features
- Module registry: 480 modules across 88 catalog categories covering browser automation, API calls, data transforms, files, crypto, notifications, triggers, queues, and workflow versioning
- Execution trace: a structured record of every step's input, output, timing, and status
- Replay: re-run from any step with the original or a modified context
- Breakpoints: pause at a step, inspect state, and resume
- Evidence snapshots: full state captured before and after each step boundary
- Data lineage: tracks data flow across steps and builds dependency graphs
- Timeout guard: workflow-level and per-step timeouts
- Verification modules: verification.* modules for site graph discovery, replay scenario generation, run evidence, and report packs
- Multiple entry points: CLI, MCP server, HTTP API, Python, and packaged recipes share one engine
- Browser automation: Playwright-based modules via the [browser] extra, including viewport changes, screenshots, and Web Vitals
- Hardened I/O: guarded HTTP clients against SSRF and a sandbox path guard on file writes (2.26.x line)
- Metering hooks: usage metering integrated into the engine
