Show HN: Seclai – See what your LLM agents are doing in production

Seclai launched on Show HN as production infrastructure for LLM agents, combining step-level observability, natural-language governance policies with flag and block thresholds, eval sets that run identically locally, in CI and in production, and switching across 200+ models from providers including Anthropic, OpenAI, Google and Meta. The platform exposes every agent as an HTTP endpoint that returns run_id, output, trace and cost on the same call, with SDKs for Python, JavaScript, Go and C#, and includes an ML-based prompt scanner that screens agent input, saved source content and webhook responses for prompt injection. Seclai positions the offering for teams that must answer for what they ship, bundling retries with fallback models, persistent memory, RAG ingestion from RSS feeds, websites, file uploads and REST APIs, and per-model, per-agent and per-step cost tracking.

Seclai is the production infrastructure for AI applications. Orchestrate workflows, evaluate outputs, enforce governance, and switch between 200+ models — visually or in code. Lead enrichmentcompletedsample and flagToday, 09:272.9s3/35.20—1 Why Seclai Built for teams who have to answer for what they ship. Every piece of a production LLM system, unified under one runtime. No orchestration stack to assemble. 01 / Observability Step-level traces Every step in every run with status, latency, credits, and governance. Replay any trace from your dashboard. Step Status Dur. Cred. Gov Input Scan Done 86ms — ✓ Evaluate response Done 1.61s 4.30 ✓ 02 / Governance Natural-language policies Write what you want screened in plain English. Every input and output is graded with flag and block thresholds you set. Content Safety Fast Flag≥0.50 Block≥0.80 PII Protection Fast Flag≥0.50 Block≥0.80 Forbidden Topics Fast Flag≥0.60 Block≥0.85 03 / Evals Same runtime for tests Define eval sets once. Run locally, in CI, and in production — with identical semantics. faithfulness: 0.94 ↑ 04 / Parity Local = production What you build and test is what runs in production. Clone any agent and it behaves identically, with nothing to re-verify or debug. support-agent → support-agent-prod ✓ workflow ✓ models ✓ policies ✓ evals 05 / Models Swap models, not code 200+ models from every major provider, switchable at agent step. 06 / Deploy Run from your app Every agent has an HTTP endpoint. POST an input and you get the output, trace, cost, and governance verdict back on the same call. SDKs for Python, JavaScript, Go, and C . POST /agents/{id}/runs X-API-Key: •••• { "input": ... } → run id, output, trace, cost How it works Three Steps to Your First Agent Go from content source to deployed agent in minutes. 01 Add a content source Enter a URL or upload a file. Seclai auto-detects the type and sets up syncing. 02 Create a knowledge base Your content is indexed into a vector database automatically for retrieval. 03 Build an agent Wire your knowledge base into a workflow with the visual builder, then deploy. Also included And everything else production demands. Built-in so you don't have to build it yourself. AI Assistant Describe what you need and the assistant configures your agent, sources, and policies. No guessing required. Resilience & Retries Auto-retry on bad output with fallback models. Structured output validation ensures quality responses. Memory Persistent conversation and general memory banks. Agents remember context across sessions. Cost Management Per-model, per-agent, per-step cost tracking. Credit-based pricing with budget controls and alerts. See It In Action The full platform, end to end. Watch a walkthrough from content ingestion to a production-ready agent with all eight pillars working together. Frequently Asked Questions What do you mean by 'everything around the LLM call'? Every production LLM feature needs more than just the API call. You need retries when output is malformed, evaluation to catch quality regressions, observability to debug multi-step pipelines, governance to screen content, cost tracking per model and user, RAG infrastructure, and persistent memory. Seclai provides all of this out of the box. How does model portability work? Select any of 200+ models from Anthropic, OpenAI, Google, Meta, and others from a dropdown. Your workflows, prompts, and data stay the same. You can also use each provider's native syntax for advanced features. What sources can I connect for RAG? RSS feeds, websites, file uploads PDFs, docs, audio, video , and REST APIs. Content is automatically chunked, embedded, and kept in sync. How does Seclai protect against prompt injection? Seclai includes a fast ML-based prompt scanner that screens all input to agents, all content saved to sources, and all webhook responses. The scanner blocks prompt injection attacks in real time before they reach any LLM, with zero LLM cost. This works alongside governance policies which screen outputs after the fact. How does governance screening work? Define safety, privacy, and compliance policies. Seclai automatically evaluates agent outputs and source content against these policies, flagging or blocking content before it reaches your users. Governance operates after the fact — the prompt scanner handles real-time input protection. Is there a free trial? Yes New accounts get a free trial with full feature access and credits to get started. The free trial is limited to one per person. Most accounts that sign up with a social login can start without a credit card. All eight production pillars are included.