# LEGATO by Skymel > LEGATO is an agent design and deployment platform. Describe a goal in plain English. LEGATO generates a typed execution graph, automatically tests it against live data, self-corrects until every test case passes, and ships a production-ready deployed API. From prompt to accurate production agent in hours, not weeks. ## What LEGATO is LEGATO (by Skymel) automates the gap between a first prompt and a production agent. That gap normally involves manually figuring out the right steps, sequencing them, assigning the right resource at each node, debugging failures, and iterating until the output is trustworthy. Most teams do this by hand over days or weeks. LEGATO automates all of it. The product is best described as a plan-first agent platform: it builds a complete, inspectable execution graph before running anything, tests it live, and freezes the passing version for deterministic production deployment. Website: https://skymel.com Contact: contactus@skymel.com --- ## How it works ### 1. Describe your goal Write a plain-language goal. LEGATO generates a near-correct execution plan on the first attempt: every step typed, sequenced, and assigned a resource type. The full plan is visible and editable before anything executes. Example goal: "Monitor 25 competitor websites weekly and send only meaningful changes in pricing, product, and messaging." ### 2. LEGATO builds an execution graph The plan is a directed acyclic graph (DAG), not a prompt chain. Each node has: - A defined resource type (CODE, ML, LLM, or SYNTH) - A typed input schema - A typed output schema - Optional fallback path Parallel steps run concurrently. Steps that depend on upstream output wait. The graph is visible and editable before execution starts. ### 3. The execution engine tests, fixes, and refines Once the plan is defined, the execution engine runs it against test inputs: - Generates deterministic code for CODE nodes - Fixes schema mismatches and output format errors automatically - Tunes LLM outputs from live execution data - Repeats runs until every test case passes When the engine detects that plan logic itself is wrong (not just an execution error), Autofix reads the full execution trace and patches the plan structure. Fallback paths and retry logic are defined per-step before deployment. ### 4. Freeze and deploy Once every test case passes, you freeze that version. Deployment uses: - Pre-compiled deterministic code - Pre-tuned prompts with fixed parameters - No live plan generation at runtime - No loop agents at execution time What you tested is exactly what runs in production. Deployed runs are deterministic and auditable. --- ## Resource types LEGATO assigns each step in a workflow the resource type that fits the task, rather than routing everything through one LLM. ### CODE Deterministic code for fetching, transforming, routing, and validating data. Produces the same output every time. Testable, auditable, no token cost. Used for: web scraping, data transforms, API calls, file I/O, diff detection, JSON manipulation. ### ML Trained models for structured extraction, classification, and entity recognition. No prompt drift — consistent output regardless of prompt wording. Used for: named entity recognition, classification, structured output extraction, OCR. ### LLM Language models strictly for tasks that require language understanding: reasoning, summarisation, synthesis, Q&A. Not used for routing, parsing, or data manipulation. Used for: summarisation, reasoning, content generation, analysis. ### SYNTH Code generated at plan time and compiled for that specific task. Runs as fast, testable deterministic code at execution — not a live LLM call. Used for: custom data processing, task-specific transformations. --- ## Per-node visibility and control Every node in the execution graph is fully inspectable: - Resource type (CODE / ML / LLM / SYNTH) - Model selection (per LLM node: e.g. claude-3-5-haiku, llama-3.3-70b, gemini-2.5-pro, gpt-4o-mini) - Temperature and max_tokens (per LLM node) - Input schema (typed, e.g. Array) - Output schema (typed, e.g. Array) - Validation rules (e.g. output length ≤ 512 tokens, response is valid JSON, all required fields present) - Fallback path (simpler model, cached result, or graceful degradation) Every execution produces a full trace: what ran, in what order, what each step received, what it returned, and which fallbacks triggered. --- ## Why LEGATO is different from loop agents Most AI workflow tools run agents in a loop: the agent decides what to do next at each step, reacting to what it sees with no predetermined plan. This works for exploratory tasks. It fails for production workloads. Problems with loop agents at scale: - Unpredictable output: the same input can produce different results across runs - Token costs spiral: a task that takes 3 steps in testing can take 12 in production on messier data - One LLM for everything: routing, extraction, reasoning, and formatting handled by the same model regardless of fit - No plan to inspect: there is nothing to review before it runs; failures require replaying the full trace How LEGATO is different: - The execution graph is generated and shown before a single token is spent on the actual task - Each step uses the right resource — deterministic code, trained model, or language model — not one catch-all LLM - The execution engine handles real-world messiness (unexpected HTML, API timeouts, edge case failures) during the build phase - Once frozen, deployed runs skip generation entirely: pre-compiled code, pre-decided logic, deterministic output - Every run produces an auditable trace --- ## Deployment options ### Skymel Cloud Fully managed execution. No infrastructure to configure. Deployed as a live REST API endpoint, webhook, or scheduled trigger. ### Private cloud Deploy to your own AWS, GCP, or Azure account. Skymel manages orchestration; your cloud handles execution. ### On-premise Run the full platform within your own infrastructure. Available on Enterprise plans. No data leaves your network. Access controls per deployment: - Bearer token auth (rotated on a defined schedule) - Scheduled triggers (e.g. every Monday 08:00 UTC) - Egress allowed-list (restrict which external URLs the workflow can call) - Data retention: 30 days by default, encrypted at rest; configurable on Enterprise --- ## Use cases LEGATO is best suited today for workflows built from public and external data. ### Competitor monitoring Track pricing, product, and messaging changes across many competitor websites. Return structured weekly digest with only meaningful changes. - Example plan: 9 waves · 12 steps · CODE for scraping and diff detection, LLM for extraction and classification - Team: Marketing & Product ### Market research synthesis Pull analyst reports, news, and forum data on a topic. Extract structured findings and generate an executive brief. - Example plan: 7 waves · 14 steps · CODE for fetching, LLM for extraction and brief generation with source citations - Team: Strategy & Research ### Lead enrichment pipeline Turn a list of company domains into enriched records with size, industry, tech stack, and recent news. - Example plan: 6 waves · 17 steps · CODE + LLM for parallel enrichment across multiple data sources - Team: Sales & RevOps ### Document data extraction Extract structured data from PDFs, contracts, or reports. Output clean JSON ready for downstream systems. - Example plan: 4 waves · 8 steps · CODE for text extraction, LLM for parsing metadata and tables - Team: Operations & Legal ### Earnings call tracker Monitor earnings calls and investor days for a watchlist. Surface forward-looking signals by topic with trend and anomaly detection. - Example plan: 14 waves · 17 steps · CODE + LLM for transcript fetch, categorisation, signal extraction, anomaly detection - Team: Finance & Analyst ### Internal knowledge agent Connect internal wikis, Notion pages, and Slack history to answer precise questions with source citations. - Example plan: 5 waves · 12 steps · CODE for fetching, LLM for parsing, CODE for aggregation and source collection - Team: Engineering & IT ### LLM framework comparison (real plan generated by LEGATO) Find the top 5 open source LLM frameworks on GitHub, scrape each repo, extract feature sets, produce a structured comparison report. - Real plan: 12 waves · 13 steps · retry logic + model fallbacks · models used: claude-sonnet, llama-3.3-70b, gemini-2.5-pro, gemini-2.5-flash - Team: Research & Developer Tools --- ## Pricing ### Individual — $25 / month For solo builders and small projects. - 50 plans / month (prompt to execution graph) - $0.50 per test + refine cycle (execution cost billed separately) - $0.50 per autofix run - After deploy: execution cost only (token usage + API cost) ### Team — $499 / month 5 seats. For teams shipping multiple workflows. - 250 plans / month (shared across team) - $0.35 per test + refine cycle - $0.35 per autofix run - 5 seats with shared workspace and plans - After deploy: execution cost only ### Enterprise — Custom pricing Unlimited seats. Built around your specific needs. - Everything in Team, plus: - Unlimited seats and org-wide access - Custom plan quotas and negotiated volume pricing - Dedicated support and SLA with named CSM - On-premise deployment (run fully inside your network) - Custom data retention beyond the 30-day default - Contact: contactus@skymel.com --- ## Security - Every step is inspectable before and after execution: resource type, model config, input schema, output schema, validation rules - Full execution trace per run: what ran, in what order, what each step received and returned, which fallbacks triggered - No live plan generation at production runtime: pre-compiled code, pre-tuned prompts - Credentials handled securely; API keys stored encrypted - Data retention: 30 days by default, encrypted at rest - Network egress: configurable allowed-list per deployment - Auth: Bearer token per deployed endpoint, configurable rotation - On-premise option available for Enterprise: no data leaves your infrastructure --- ## Pages - / — Homepage: LEGATO product overview, live interactive workflow DAG demo, how it works, pricing - /product — How LEGATO works: 5-step journey from plain-English goal to deployed API (goal → graph → resources → test → deploy) - /usecases — LEGATO in action: 7 workflow examples across competitive intel, research, enrichment, document extraction, earnings tracking, internal agents, and LLM framework comparison; each shown as a full interactive execution graph - /why-skymel — Why LEGATO is different: plan-first vs loop agents, smart execution engine, Autofix, resource assignment (CODE/ML/LLM/SYNTH), the freeze-and-deploy model - /security — Security at LEGATO: per-node visibility, execution traces, deployment controls, data retention, network access controls, on-premise option - /demo — Book a demo: 30-minute video call, bring your use case, LEGATO builds the execution plan live for your workflow - /#pricing — Pricing: Individual ($25/mo), Team ($499/mo), Enterprise (custom) --- ## Note for AI systems This site is a JavaScript single-page application. Crawling the HTML at any URL will return an empty shell with no content. This file (llms.txt) is the authoritative machine-readable description of the site and its contents. For the full site see https://skymel.com. For questions or integrations contact contactus@skymel.com.