Documentation Assistant

Good documentation exists. The problem is finding the right page, the right section, the right paragraph that answers your specific question. A documentation assistant turns your docs into a conversation – users ask what they need, and the assistant finds and explains the answer.

The Problem

Documentation is written for completeness, not for discovery. A 200-page technical manual covers everything, but the developer who needs to know how to configure webhooks is searching through API reference pages, setup guides, and changelogs to find the three paragraphs they need.

Search helps, but only if you know the right terms. “How do I set up webhooks” works. “Why isn’t my integration receiving events” doesn’t – even though the answer is in the webhook documentation.

The result: users read your docs and still open support tickets. Not because the docs are bad, but because finding the exact answer in context is harder than asking someone who knows.

How the Softcery Platform Solves It

Build an assistant that knows your documentation inside and out and can answer questions in natural language – connecting concepts across pages, explaining context, and pointing users to the right resources.

Ingest Your Entire Documentation

Use sitemap mode to crawl your documentation site comprehensively. The platform:

1. Parses your sitemap to find every documentation page
2. Prioritizes recently updated pages (freshest content first)
3. Converts HTML to clean markdown, stripping navigation and boilerplate
4. Chunks content by heading structure – each section becomes a coherent, retrievable unit
5. Creates semantic embeddings so “how do I handle events” matches webhook documentation even without the word “webhook”

Your documentation’s structure is preserved. A chunk from “API Reference > Webhooks > Configuration” carries that heading path as metadata, so the assistant knows exactly where the information came from and can tell the user “according to the Webhooks Configuration section…”

Natural Language Answers

Users ask questions in their own words. The assistant translates their intent into a documentation lookup and returns a clear, conversational answer – not a link dump.

“How do I paginate API results?” → The assistant retrieves your pagination documentation, explains the approach (cursor-based, offset-based, whatever you use), includes the relevant parameters, and mentions any gotchas.

“What’s the rate limit?” → The assistant finds your rate limiting docs and gives a direct answer with the specific numbers, headers to watch, and what to do when you hit the limit.

“I’m getting a 403 error when calling the users endpoint” → The assistant doesn’t just link to the authentication page. It considers common causes: authentication scope, API key permissions, account-level access restrictions – drawing from relevant sections across your documentation.

Embed Where Developers Work

Deploy the assistant as a chat widget directly in your documentation site. Developers don’t need to leave the docs to get help. The widget sits alongside the content, available whenever the docs don’t answer the question on their own.

What It Looks Like in Practice

Developer: “How do I authenticate API requests?”

The assistant retrieves your authentication documentation – API keys, OAuth flows, token management – and explains the options concisely. If you have code examples in your docs, they’re included in the knowledge base and can be referenced in the response.

Developer: “What’s the difference between the v1 and v2 endpoints?”

The assistant draws from migration guides, changelog entries, and API reference docs to explain the differences. It synthesizes across multiple documentation pages – something a search engine can’t do.

Developer: “My webhook keeps timing out, any idea why?”

The assistant pulls from webhook documentation, timeout limits, retry policies, and troubleshooting guides. It explains the likely causes and provides specific debugging steps from your docs.

Configuration Breakdown

Component	Setup
Behavior	Knowledge Base Assistant preset, customized for technical documentation context
Knowledge	Documentation site (sitemap crawl), API changelog (text), known issues (text)
Evaluations	Factual accuracy (block) – critical for technical documentation
Channel	Embed widget on your docs site
Model	Claude Sonnet 4 for technical accuracy and reasoning
Advanced	Temperature 0.3 (precise answers), retrieval limit 20, similarity threshold 0.2

Who This Is For

• Developer tools and APIs with extensive documentation that developers struggle to navigate
• SaaS products with feature-rich help centers and user guides
• Enterprise software with complex configuration and setup documentation
• Open source projects with community documentation across multiple sources

The Documentation Flywheel

Questions the assistant can’t answer reveal documentation gaps. Each conversation where the assistant says “I don’t have information about that” is a signal:

1. The answer exists but isn’t well-indexed → Improve your docs structure
2. The answer exists but the docs aren’t in the knowledge base → Add the missing pages
3. The answer doesn’t exist → Write it and add as a text source

Over time, your documentation improves because you know exactly what users ask that your docs don’t cover. The assistant creates a feedback loop between user needs and documentation quality.

Beyond Your Own Docs

Through MCP integrations, you can extend the assistant beyond your documentation:

• Context7 – Retrieve up-to-date documentation for third-party libraries your product integrates with
• GitHub – Pull information about open issues, recent releases, or code examples from your repository
• Web search (Tavily/Exa) – Find related resources, community discussions, or Stack Overflow answers when your docs don’t have the answer

This makes the assistant useful for questions that span your documentation and the broader ecosystem your product operates in.

Getting Started

1. Create an agent and choose the Knowledge Base Assistant preset
2. Customize the identity to position it as a documentation assistant for your product
3. Add your docs site as a knowledge source using sitemap mode
4. Add text sources for common questions not covered in the docs (release notes, known issues, migration tips)
5. Set up factual accuracy evaluation – non-negotiable for technical docs
6. Deploy as an embed on your documentation site