Roadmap - Featurebase

Backlog

61

Under consideration

@mention Fibi in Slack to get answers

Allow customers and teammates to @mention Fibi in a Slack channel or thread to get answers from your documentation. Customers could ask Fibi before tagging your team. Teammates could bring Fibi into an existing discussion to help answer a customer's question or find information internally. Fibi should use the thread for context and reply in the same place, with a way to hand over to your team when human help is needed. Quick internal questions should not create unnecessary support conversations.

Contradiction checker for Fibi training content

Teams should be able to check existing Q&A, uploaded files, crawled websites, and Help Center articles for potential contradictions, on demand or on a schedule. Results should show the conflicting passages and links to both sources for review, without requiring a proposed Q&A.

Fibi launches tours for how-to questions

Enable Fibi to launch a relevant product tour when a user asks a how-to question, so tours serve as guided answers instead of text-only replies.

Advance notice before AI model changes

Notify customers in advance before the underlying AI model behind the support agent is changed, including the lead time and the channel used, and disclose which AI model(s) are currently in use. Prompts and knowledge bases tuned to the current model's behavior can degrade in answer quality when the model changes without warning, making performance changes hard to trace. The primary use case is investigating changes in answer quality rather than gating rollout approvals, but advance notice remains a desired feature. As a possible approach, the data science team evaluating the product suggests handling model changes like software versioning: release a new version of Featurebase that uses the different model while still supporting the current version, let customers evaluate and tune their prompts for the new model, then switch to the new version, after which support for the old version ends.

Manage Fibi custom actions via API & MCP

Create and update Fibi custom actions through the API and MCP. This would let an AI agent configure and maintain actions without having to complete each setup through the dashboard.

Configurable crawling interval and manual force recrawl for Fibi training data

Allow setting custom automatic recrawling intervals (e.g., daily or multiple times per day) for website training data sources like status pages, along with a direct button to force an immediate recrawl.

AI conversation attribute categorization without Fibi as a step

I want AI conversation attributes to categorize and route conversations based on their meaning, without requiring a 'Let Fibi answer' step or a reply to the customer first. Exact keyword matching is not reliable enough. For example, we route incoming emails to either our Questions or Bugs team. Questions should use our documentation to answer the customer. Bugs need diagnostic questions to collect the missing details, then a faster handoff to a human. The AI should first read the email silently, set the relevant conversation attributes, and let the Workflow route it to the correct team and instructions. Fibi can then reply if that branch requires it, or the conversation can go straight to a human.

Improve Fibi's resistance to completing unrelated tasks

Improve Fibi’s resistance to use as a general-purpose assistant. Fibi should decline unrelated tasks, such as translating personal messages, and redirect users to company support topics. Visitors can currently abuse Fibi as their GPT alternative in certain scenarios.

Toggle to allow or disallow images in Fibi responses

Add a setting to control whether Fibi can include images in responses. When disabled, Fibi must send text-only replies, including no separate image galleries or email image attachments.

Option for AI Agent to look for in a specified collection of articles for a workflow

If i have 3 Workflows for A, B, C, i would like Fibi AI agent to only look for in the corresponding collection of article. If the workflow A is running, Fibi AI agent should only look in the A collection articles, etc. Today he is mixing collection in this anwser leading to totally inconsitent answer. We’ve tried to be strict in his prompt but it still happen sometimes.

Planned

4

Committed and queued

Fibi follow-up setting customization

There should be a way to configure Fibi's follow-up behavior, including options like: Fibi will confirm whether the user still needs assistance - Fibi asks a short follow-up question after answering, such as “Did that answer your question?” Fibi will offer escalation - Fibi checks in and offers to connect the user with a teammate if they still need help. No follow-up - Fibi sends only the main answer and does not send follow-up prompts or idle nudges. Fibi, currently always sends follow-up prompts and idle nudges based on internal logic, but there is no surfaced setting to disable or customize that behavior.

AI-detected conversation tags

Allow Fibi to automatically set conversation tags based on conversation context, similar to how conversation attributes can currently be detected and updated by AI. Today, tags can be applied through workflow rules such as keywords or explicit tag actions, which does not cover cases where the correct tag depends on understanding the full conversation.

Support Messenger User Context

The ability to provide the support messenger AI with additional contextual data about the user. This could be taken from their user attributes or passed to the support module at the time of sending a message. User Cases The user does/doesn’t have a subscription and therefore the AI can respond apprioty when a user asks about a paid feature (not saying they already have it when they don’t and vice versa). Knowing which page the user is currently on and maybe even more specific details. A user will often ask about something they are currently looking at and the AI should be aware of that. A user might be using an old version of your software, the AI should know this so that debugging can be aware that the problem has already been fixed in a newer version.

Automatic article gap detection for Fibi AI

Scenario Sometimes Fibi can’t answer a of question specific topic due to Help Center article not existing for that topic. If many similar un-answerable questions accrue, then Fibi should be able to create a Request for an Article for given topic, so it can be more helpful in the future. Solution Unanswered topics section in the Inbox with an option to Create an article for the topic Fibi will convert unanswered questions, that seem relevant to the product, into topics that will be collected into the Unanswered topics section

Next up

2

Committed and queued

Fibi AI Table Parsing in Help Center

I want the AI agent (Fibi) to be able to read and extract information from tables in the help center. Currently, it seems to only process plain text, which means important data stored in tables is missed during AI searches and chat interactions. This is important because many of our technical specifications and pricing details are in tables.

Have Fibi index content from Help Center's FAQ elements

Currently, Fibi (and probably Help Center AI) doesn’t index and use info from the Help Center accordion elements used for FAQ. It’s super important to get this fixed, as we’re getting questions that are not answered but are documented.

In Progress

1

Actively being built

Internal articles and snippets for Fibi and Copilot

Teams need to give Fibi and Copilot detailed context that should not be published as Help Center content. Today this knowledge is managed separately in Fibi Training Data and does not behave like first-class, reusable content. Add dedicated internal content types to the shared Knowledge area, similar to Intercom: Internal article: long-form, richly formatted knowledge for procedures, policies, product details, troubleshooting, and other material that should remain private. Internal snippet: short, reusable facts or instructions that do not need a full article. For every internal article or snippet, let teammates independently enable or disable whether it can be used by: Fibi AI Agent Copilot Other AI or support surfaces added in the future Internal content must never appear publicly in the Help Center. It should still use the same high-quality editor and management model as public Help Center articles, rather than the current separate Training Data format. Internal articles and snippets should be searchable, filterable, organized, and show useful metadata such as where they are used, owner, language, and last updated date. They should also be fully manageable through the API and MCP. This is one part of the broader unified knowledge management for Help Center and Fibi AI Agent direction.

An Anonymous User

Done

23

Recently shipped

Reliable bulk upload of Fibi training files

Uploading more than two training files at once in Dashboard → Automations → Training Data is unreliable. Files fail very quickly and land in a Failed state, even for small files (for example a 1.8 KB Markdown file that reported 0 words). Failed files then have to be removed one by one and re-uploaded, which makes seeding an agent with a real documentation set slow and error prone. Requested behaviour Multi-file selection should upload reliably, with queueing or throttling instead of parallel failures. Show the actual reason a file failed (size, type, conversion error, rate limit) rather than a generic Failed label. Offer a retry action on failed files instead of requiring manual delete and re-upload. Reported by a customer evaluating Featurebase who is loading a Markdown documentation set into the AI agent.

Fibi should not respond to every Slack message

Fibi can reply to Slack messages even when they aren't questions and no assistance is needed. In one example, it replied to a message explicitly saying a response wasn't wanted, then replied again to a short “See” follow-up. Fibi should recognize when to stay quiet instead of engaging with every message. This would reduce unnecessary bot replies and let normal customer conversations continue naturally. When uncertain, it can lean toward answering, but it should skip messages that clearly don't need a response.

Improve Fibi accuracy, reliability & speed

Rework customer support AI Fibi to be more reliable and fast.

Exclude specific modules (Changelog/Feedback) from Fibi AI training

I want to be able to exclude specific modules (like the Changelog) from Fibi's training data. Currently, all changelogs are automatically indexed, which can lead to outdated information being used in AI responses. Having a toggle to disable certain modules or specific posts from training would help ensure more accurate and refined answers.

Reduce AI Agent Latency

I want the AI agent to respond much faster. Currently, the latency is too high, which creates a poor experience for users. This seems to be a general issue rather than account-specific.

Train/correct Fibi AI on its existing answers

Allow teams to select existing AI answers and use them as training signals for future responses by clicking “Improve answer” and inserting the ideal answer. This would let teams tell Fibi “this is the correct answer” and have it respond similarly next time, without needing to update help articles or add separate guidance.

Setting to Control Fibi's auto-solving (auto closing) and conversation resolution

I would like to turn off the auto-solving questions from Fibi AI. Specifically, I'd like for Fibi to leave questions open until a human manually closes the conversation, not marking anything as resolved automatically. Additional improvement would be to customize the timer before closing the conversation as well.

Fernando Santos

AI conversation attribute categorization

Use Fibi AI to automatically classify incoming conversations into defined attributes - so you can route faster and act smarter. Automatically fill in custom fields like product, priority, or issue type based on conversation content with AI to save time for human agents.

Internal AI reference codes are visible to end-users in support messages

As a user, I am seeing internal reference codes, like [REF-RESEARCH:XYZ], in the support messages from the AI bot. These codes should not be visible to end-users as they are confusing and serve no purpose for them. This seems to be a bug in how the AI agent's internal markers are processed and displayed.

Fibi isn't asking for email on hand-over & when anonymous users try to submit Posts

When anonymous posting is disabled Fibi AI should ask for email before allowing to submit a feature request / a feedback post - currently it isn’t asking for email and allows to directly submit Posts without identification It should be asking for an email on hand-over as well