Seclai

Seclai: build, run and govern AI agents over your own content

Seclai is a platform for building AI agents, chat bots and MCP servers on top of knowledge bases that keep themselves current. Connect your content, design a multi-step workflow, pick from a wide catalog of models, and ship it to users, to your own code, or to other AI tools. Governance, prompt-injection scanning, evaluations and full run traces are built in, not added later.

KNOWLEDGE BASES THAT STAY CURRENT
Point Seclai at websites, RSS feeds, Google Drive or Dropbox folders, inbound email, or uploaded files. Content is pulled on a schedule, extracted, chunked and indexed for retrieval. Audio and video are transcribed, and images on web pages, in feeds and inside documents are indexed alongside the text, so retrieval works across media and not only on plain text. Metadata filters and reranking keep results focused on what a step needs.

AGENT WORKFLOWS, NOT JUST PROMPTS
Agents are workflows made of typed steps that you build in the editor or define as JSON. Steps cover retrieval, LLM prompt calls with tools, web search and fetch, data extraction (JSON, HTML, XML and structured data), branching (if/else and switch), loops over runtime lists, calls to other agents, calls to tools on external MCP servers, memory reads and writes, waits, and human-in-the-loop approval that pauses a run until reviewers decide. Outputs can go to webhooks, email, S3, back into a knowledge base, or stream straight to the caller. Image, audio and video generation steps are available too. An AI assistant can draft a workflow from a plain-language description to get you started.

Agents run when new content arrives or changes, when a file is added to a connected cloud drive, when an email reaches the agent's own inbox, or on demand through the API.

MODEL CHOICE WITHOUT LOCK-IN
Choose models from Anthropic, OpenAI, Google and many other providers per step, with effort and reasoning controls where a model supports them. Seclai tracks model deprecations and sunset dates, warns accounts that use affected models, recommends successors, and can upgrade a step automatically with optional rollback. Requests can fail over across vendors that host the same model.

GOVERNANCE AND SAFETY BUILT IN
Content is screened where it enters: agent input, external data returned by a step, and new content arriving in a source. Define governance policies with configurable evaluation tiers, hold content for human review, and keep an audit trail of every policy change. A dedicated classifier scans inputs for jailbreak attempts. Inbound email is gated by sender allowlists backed by DMARC, not just a bare signature check.

QUALITY YOU CAN MEASURE
Agent evaluations score step outputs with an LLM judge, either on every run with automatic retry, as sampled monitoring in production, or as manual validation. Evaluations can run retroactively over past runs. Every run produces a step-by-step trace that streams live while the agent is working, including inputs, outputs, tokens and credit cost for each step.

MEMORY AND CONVERSATIONS
Named memory banks give agents durable facts and per-conversation chat history with vector search and automatic compaction, so long-running assistants keep context without blowing up prompt size.

BUILT FOR DEVELOPERS
Agents, knowledge bases and runs are available through a versioned REST API, with official SDKs for Python, JavaScript/TypeScript, Go and C#, plus a CLI. Seclai also runs as an MCP server, so tools such as Claude and ChatGPT can query your knowledge bases and run your agents directly, using OAuth or API keys. Agent definitions export as JSON for version control and import into another account or environment, and API versioning protects existing integrations as the platform evolves.

FOR TEAMS AND LEADERS
Organizations share agents, knowledge bases and connections across members. Usage is metered in credits with per-run and per-step cost visibility, low-credit alerts and automatic