If you’ve ever asked a support bot a slightly unusual question and gotten a completely irrelevant answer, you already know the problem with keyword-based FAQ matching. The customer asks “what’s the policy on remote work during my first three months?” and the bot responds with a generic attendance policy because it matched on the word “policy.” The nuance — probation period, remote work specifically — gets lost.
Today we’re shipping two improvements to JieGou chat agents that address this directly: Embedded FAQ Mode for more comprehensive answers, and thread isolation for cleaner conversations across group chats and DMs.
The problem with keyword matching
JieGou chat agents have always supported FAQ automation through Rules mode — you define keywords and patterns, and the agent responds with the matching answer. This works well for straightforward questions: “what are your business hours?” matches the keyword “business hours” and returns the right answer every time.
But real customer and employee questions are rarely that clean. People ask compound questions. They use different phrasing. They reference context that doesn’t appear in your keyword list. Rules mode handles the common cases, but the long tail of questions often falls through to human escalation — even when the answer exists somewhere in your FAQ document.
Three FAQ modes, one agent
Chat agents now support three FAQ modes. You pick the one that fits your use case.
Rules (default). The original keyword and pattern matching. Fast, predictable, and low cost. Best for agents handling a small set of well-defined questions where exact matching is sufficient.
Embedded. The entire FAQ document is injected into the LLM’s context window for every incoming message. The model reads your full document and generates answers based on comprehensive understanding — not keyword overlap. This means it can handle rephrased questions, compound questions, and questions that require synthesizing information from multiple sections of your FAQ.
Hybrid. The agent tries Rules first. If no pattern matches, it falls back to Embedded mode and sends the full document to the LLM. This gives you the speed and low cost of keyword matching for common questions, with the comprehension of full-document context for everything else.
Under the hood, Embedded and Hybrid modes use loadFAQDocuments() to fetch and cache your FAQ content with a 5-minute Redis cache. The document is loaded once and reused across requests, so you’re not paying for redundant fetches.
A concrete example: the HR onboarding bot
Consider an HR team that deploys a chat agent on Slack to answer new-hire questions. The agent’s FAQ source is the full employee handbook — 40 pages covering benefits, policies, IT setup, and office logistics.
With Rules mode, the team would need to manually extract every possible question and map it to a keyword pattern. “Remote work” → section 4.3. “PTO” → section 7.1. Any question that doesn’t match a predefined pattern gets escalated.
With Embedded mode, a new hire can ask: “I’m in my first month — can I work from home on Fridays, and do I need to submit a request in advance?” The LLM reads the full handbook, finds the probation-period policy in section 2.4 and the remote work request process in section 4.3, and synthesizes a single coherent answer. No keyword mapping required.
The Hybrid approach works well here too: common questions like “where do I pick up my laptop?” match instantly via rules, while the unusual questions get the full-document treatment.
Thread isolation: group chats and DMs stay separate
The second improvement is about conversation reliability. When a chat agent operates in both group chats and DMs on the same channel — which is common on LINE, Slack, Discord, and MS Teams — the conversation threads are now fully isolated.
Previously, messages from a group chat and a DM with the same user could share thread context, which occasionally caused the agent to reference group chat discussions in a private DM response, or vice versa. This was confusing at best and a privacy concern at worst.
Thread isolation works by incorporating the group or room ID into the conversation scope. A message in “Marketing Team” group chat and a DM from the same user are now tracked as completely separate conversation threads. The agent maintains independent context for each, and there’s no cross-contamination.
This applies across all 12 messaging channels that JieGou supports: LINE, Instagram, Facebook Messenger, WhatsApp, YouTube, Slack, Discord, Microsoft Teams, SMS, email, web chat, and Telegram. If the channel supports both group and direct messaging, thread isolation is enforced automatically.
Getting started
Both features are available now for all JieGou accounts. To switch your chat agent’s FAQ mode, go to the agent configuration page and select Rules, Embedded, or Hybrid under the FAQ section. Thread isolation is enabled by default — no configuration needed.
If you’re running FAQ-heavy agents on support, HR, or internal help desk channels, try Embedded or Hybrid mode. The difference in answer quality for nuanced questions is significant.