HAI Assistant is a conversational AI agent built directly into your Harvestr workspace. It connects to your feedback data (feedback, discoveries, components, customers) and lets you query, analyze, and summarize it through natural language.
Rather than navigating filters and views manually, you can ask HAI Assistant directly and get structured, cited answers grounded in your actual data.
HAI Assistant is available on all paid plans. It is distinct from HAI's automated feedback processing features: learn more about HAI AI-powered feedback processing →
Usage limits
Every plan includes a monthly allowance of AI asks.
Plan | AI asks per month |
Core | 100 |
Scale | 250 |
Elite | Custom |
What counts as one ask? One question you send to HAI, plus its full response. That response can be a quick answer or a substantial piece of work: analyzing your feedback, drafting a spec, updating your product data. However long it takes, it counts as one ask.
Asks are pooled across your whole workspace and shared by editors and contributors, whether they ask in Harvestr or in Slack. You can track this month's AI asks in Settings > Billing, in the Usage card. Full details on the pricing page.
Where to access HAI Assistant
HAI Assistant is available in three places:
Main chat: accessible from the left navigation bar, this is the full conversational interface where you can ask questions across your entire workspace, or focus them by adding context.
Discovery tab: each discovery includes a dedicated HAI Assistant tab. Open any discovery and click the HAI tab to ask questions scoped directly to that discovery (its feedback, sub-discoveries, and linked data) without any additional setup.
Slack: once the Slack integration is connected, mention @Harvestr in a channel to ask a question (HAI answers in a thread), or send it a direct message.
Using HAI Assistant
Starting a new conversation
Open HAI Assistant from the left navigation bar. You'll land on the main chat screen with an input field at the bottom, and your recent chats below it under Pick up where you left off.
To start a fresh conversation, click the pencil icon in the top right of the chat panel. To reopen a previous conversation, click the clock icon next to it, or All sessions under your recent chats.
Asking a question with context
Without context, HAI Assistant searches across your entire workspace. To focus its analysis, click Add context below the input field, or type @ in your message, then pick what you want to talk about:
A component: analysis focuses on the discoveries and feedback within that component
A discovery: HAI Assistant focuses on that discovery and its sub-discoveries
A user, a company or a teammate: HAI Assistant takes that person or account into account in its answer
This is especially useful when you want a focused deep-dive on one area.
For example, adding your "Billing" component before asking "What are the main pain points?" will return a more relevant answer than asking globally.
Note: when accessing HAI Assistant from a discovery tab, the scope is automatically set to that discovery.
Viewing sources
After HAI Assistant responds, a Sources button appears below the answer and in the top right of the chat panel. Click it to open the sources panel, which lists all the feedback items and discoveries referenced to produce the answer, organized into Feedback and Discoveries tabs.
Each source links directly to the original resource in Harvestr, so you can verify any claim or read the full feedback behind a cited extract.
When HAI has made changes in your workspace with AI actions, an Applied changes button lists them, with Compare changes to review what was modified.
Note: the Sources panel is not available when using HAI inside of a Discovery.
What HAI Assistant can do
Check whether you already have feedback on a topic
Ask whether feedback exists on a topic and HAI Assistant answers with one exact number, then shows where it lives: the discoveries that cover the topic, each with its feedback count and state, a few customer quotes, and the matching feedback not linked to those discoveries yet. The count includes suggestions still waiting for review in Catchup, and says how many there are.
Example queries:
"Do we have feedback on SSO?"
"Is there a feature request for better reporting?"
"Which discovery covers payroll exports?"
Summarize feedback for a discovery or topic
HAI Assistant pulls together all feedback linked to a discovery and surfaces the key themes, pain points, and patterns, saving you from reading through dozens of individual pieces of feedback manually.
Example queries:
"Summarize the feedback on the Dashboard performance discovery"
"What are customers saying about our onboarding flow?"
"What are the main pain points mentioned in the Billing component?"
Find feedback from a specific customer
HAI Assistant retrieves all feedback submitted by a given user or company: useful for preparing customer calls, understanding churn risks, or identifying your most vocal accounts.
Example queries:
"Summarize all feedback from Acme Corp"
"What has Acme asked for over the past 6 months?"
"Which customers have mentioned our API integration?"
List the feedback or customers behind a topic
When you explicitly ask for a list, HAI Assistant shows one page of up to 25 items with the total, each linked to Harvestr. Ask for the next page to see more. To see which companies ask most, ask for a ranking instead.
Example queries:
"List the feedback on the Export discovery"
"Which companies asked for SSO?"
"Who submitted feedback on this discovery?"
Segment by customer attributes
HAI Assistant can read the customer attributes synced into Harvestr (plan tier, ARR, industry, geography, and so on) and use them to segment its analysis.
Example queries:
"What are Enterprise customers asking for that others aren't?"
"Summarize feedback from our French customers"
"Which discoveries matter most to accounts above 50k ARR?"
"How many Enterprise customers asked for SSO?"
Identify trends and volume changes over time
HAI Assistant analyzes feedback volume across a time period and surfaces spikes, drops, or seasonal patterns, including a breakdown by discovery to show which topics are growing fastest.
Example queries:
"How has feedback volume evolved over the last quarter?"
"Which discoveries received the most feedback in January?"
"Are there any topics that spiked in the last 30 days?"
Explore your discovery backlog
HAI Assistant can search, filter, and rank discoveries by state, component, feedback volume, or the values of your custom fields: scores such as RICE or effort, Rollup fields, select fields, text and checkbox fields.
Example queries:
"What are my top 5 discoveries by number of feedback items?"
"List all discoveries currently in the 'In Progress' state"
"Which discoveries in the Analytics component have no linked feedback?"
"Show the 10 discoveries with the highest RICE score that have an effort of 2 or less"
Take actions in your workspace
With AI actions, HAI doesn't just answer: it does the work. It can do most of what you can do in Harvestr: draft a PRD, update a feature's status or its custom field values, merge duplicates, reprioritize your backlog.
Available from the Scale plan. Each action counts as one AI ask.
Example queries:
"Draft a spec for the Notifications discovery based on its feedback"
"Merge the duplicate discoveries about SSO"
"Move everything released last sprint to the Released state"
Support strategic prioritization
On strategic queries, HAI Assistant synthesizes findings into recommendations grounded in your data. Every suggestion traces back to cited feedback, not generic best practices.
Example queries:
"Based on recent feedback, what should I focus on this quarter?"
"What are the biggest unresolved pain points across all open discoveries?"
Answer operational questions instantly
HAI Assistant handles direct, factual lookups without requiring you to build a view or apply filters.
Example queries:
"How many discoveries are currently in the Backlog state?"
"Which discoveries were shipped last month?"
"How many feedback items did we receive from Zendesk in Q1?"
How to get the best results
Be specific about scope. Include the discovery name, customer, or time period in your query. Broad questions without constraints may return partial results or prompt HAI Assistant to ask for clarification.
Less effective: "What's happening with feedback?"
More effective: "What feedback did we receive on the Reporting component in the last 60 days?"
Add context to pre-scope your question. If you know you want to focus on a specific component, discovery or customer, add it with @ before sending. This reduces noise and improves response quality.
Provide a time range for broad searches. When searching across your entire workspace without a specific filter, HAI Assistant will ask for a date range. Having one ready speeds things up.
Use follow-up questions to drill deeper. HAI Assistant supports multi-turn conversations: you can build on previous answers without starting over. For example:
"Summarize feedback on the Notifications discovery"
"Which customers mentioned email specifically?"
"What are those customers' specific needs?"
Bundle related work into one ask. Since an ask is counted per question and not per amount of work, a single well-scoped request that produces a full analysis costs the same as a one-line answer.
Current limitations
Harvestr data only: HAI Assistant works exclusively with data in your Harvestr workspace. It reads the customer attributes synced into Harvestr, but has no direct connection to your CRM, and cannot browse external sources.
No memory between sessions: Each conversation starts fresh. Context from a previous chat is not carried over automatically.
Text-based feedback only: HAI Assistant cannot analyze attachments or images.
No exports: HAI Assistant shows lists one page at a time and can't produce a file. To get every row, export your feedback.
Looking for an AI tool that works from outside Harvestr? Our MCP server connects Claude, Cursor, and other assistants to your workspace.
AI models and data
HAI Assistant, in Harvestr and in Slack, runs on Anthropic's Claude models. Using it means accepting Anthropic as a subprocessor of your data. Your data is not used to train any model.
The full list of subprocessors and our security policies are in our Trust Center.
Sharing feedback on HAI Assistant
Your feedback directly shapes how HAI Assistant improves. If a response is incorrect, incomplete, or misses your intent:
Tap the 👎 icon below any response to flag it
Share details in the support chat via the chat bubble in the bottom right of the screen



