Skip to content
Evergrowth Documentation

Eva

Eva is the one colleague in the workspace you talk to directly. Instead of deciding which specialist agent to run, in what order, and on which records, you describe what you are trying to get done in plain language, and Eva works out which of your AI colleagues to bring in and runs them for you.

What she is

Section titled “What she is”

Every other AI colleague in Evergrowth is a specialist: one job, done well. Eva is different in kind. She is the conversational entry point to the whole workspace - the one you brief the way you would brief a sharp colleague, rather than a screen you configure. She reads the same shared intelligence every agent draws from - your value proposition, your verticals, your personas, your priority accounts - so she already has the context to reason about your business before you explain anything.

Because she can act inside the workspace, not just describe it, a conversation with Eva can turn directly into work: she can look at what is already true about an account or a contact, kick off the research or qualification that is missing, and adjust pieces of your shared setup on request - the same way you would ask a colleague to make a change, rather than clicking through it yourself. See the Agent Training Center for what that shared setup contains.

Why she exists

Section titled “Why she exists”

Most of the product is organized the way a RevOps lead thinks about a GTM system: by capability, one agent per job. That is the right shape for building and for precision, but it is not the shape a rep or a manager thinks in when they sit down with a real question - “is this account worth my time,” “what’s actually going on with this deal,” “where should I focus this week.” Those questions cut across several agents at once, and answering them well means knowing the org chart of specialists well enough to pick the right one.

Eva removes that translation step. You do not need to know that “is this worth my time” means running qualification, then research, then checking for a signal - you ask the question, and she works out what to check and in what order, the same way a capable colleague would rather than handing you a manual.

What she is not

Section titled “What she is not”

Eva is a strategist and an orchestrator, not the agent that produces a specific deliverable. That distinction matters for what to expect from a conversation with her:

  • She does not write your outreach herself. Drafting an email, a call script, or a message is Play Copywriting’s job. Ask Eva about strategy - whether an account is worth pursuing, what to do first - and she will either explain her reasoning or hand off to the agent that produces the actual copy, rather than writing it in the conversation.
  • She is not the Digital Twin. The Digital Twin is a different conversational colleague that lives on one specific contact, for rehearsing how a single call might go - positioning, objections, what this buyer specifically cares about. Eva operates at the level of the workspace and your whole book of accounts: prioritization, setup, and cross-account reasoning, not a rehearsal partner for one conversation.
  • She is not a substitute for playbooks. For repeatable, standing automation - a workflow that should run the same way every time a new account comes in - playbooks are the right building block, not a recurring chat with Eva. Reach for Eva when you have a question or a one-off judgment call to make; reach for a playbook when you want the same sequence to run every time without asking again.

How she works in a conversation

Section titled “How she works in a conversation”

Eva is meant to behave like a capable colleague rather than a search box, and a few of those habits are worth knowing because they change how you brief her.

  • She acts first and confirms after. Asked for something with an obvious sensible reading - “give me my top ten accounts” - she goes and gets it rather than asking you what you meant. She ranks the result by the most defensible measure available, tells you in one line which measure that was, and offers the other readings as alternatives you can pick from. She holds back and asks only when the possible readings mean genuinely different work and none of them is a fair default.
  • Answers come back ranked, with a named first move. A list of accounts, people, or fixes arrives in priority order with one reason each and an explicit recommendation of what to do first, so you are not handed a pile to triage yourself.
  • She leads with what you did not ask about but need to know. A contact who has left the company, a newly vacant seat worth being early to, a shift in what an account sells that changes why they would buy: that comes up front, not as a footnote after the thing you asked for.
  • She checks before telling you something is missing. If one place looks empty she looks at another that would show the same thing, and trusts the actual records over any single view. If she has already told you something was not there and then finds it, she says so and corrects herself rather than quietly moving on.
  • She looks for the cause, not the symptom. When something is blank, missing, or will not sync, she goes one level deeper before acting, fixes what actually caused it, runs it again, and tells you in a line what was wrong - instead of re-running the same thing and handing you the same empty result.
  • When she cannot do something, she says so once and still moves you forward. You get the limit in a sentence, then the next best thing in the same reply: the manual route, and the content already written so you only have to paste it. She does not end on the obstacle or promise to come back to it later.
  • You can show her something, not only describe it. A screenshot carries what a paragraph of description usually loses: the actual chart, the exact error, the slide a prospect sent, the page you are looking at. Attach one to your message - by picking it, dragging it in, or simply pasting it - and she reads it as part of the question, so “why does this look wrong” is a question you can ask about the thing itself. Up to five images travel with one message, and an image with no words is a perfectly good message. She works in the same way wherever you are talking to her, so a screenshot shared in a chat thread reads the same as one attached in the workspace. Files that are not images are a different matter: she says she could not read them and answers on what she does have.
  • Her answers are laid out, not typed out. A set of records comes back as a table you can scan rather than a paragraph listing them, a single record as something you can open directly, and the sensible follow-ups as choices you can select instead of questions you have to retype. A pointer into your workspace is always something you can actually open - she does not write out a made-up address and leave you to discover it goes nowhere.
  • A long piece of work says how long it has been working, and survives you leaving. Some questions take real work - reading a book of accounts, running research before answering. Past the point where a pause stops reading as thinking, she tells you how long the turn has been going and gives you the option to stop it. You can also close the page or go elsewhere in the workspace: the work carries on and the conversation catches up when it is done, so a long answer does not depend on you sitting and watching it.
  • A limit she ran into is stated as information, not as a failure. There is a ceiling on how much she will do inside a single turn, so that one question cannot quietly consume an afternoon’s worth of work. When she reaches one, she says what she covered and where she stopped, in the neutral tone of a status note rather than an error - because the work she did is still good, and the honest next step is usually to narrow the question or ask her to continue. When something genuinely does break - a step that took too long, a system that did not answer - she names what happened instead of shrugging, so you can tell “this timed out, try again” apart from “this cannot be done”.

What she costs

Section titled “What she costs”

Eva draws on the same credit pool as every other AI colleague, priced for the work she actually does rather than for talking to her. Asking a question is free; the pieces of real work inside her answer - reading your accounts, searching for people, running an agent - each draw a small amount, and the workspace shows you the price beside each step and the total under the finished answer. Nothing is charged for her laying the answer out, for a step that failed, or for a step that ran and found nothing, so the prices you can see are the whole cost of the turn.

That is worth knowing mainly because it changes how it pays to brief her. A sharper question is a cheaper one, not because the conversation is metered but because it takes less looking around to answer - and a broad “tell me everything about my book” costs what the work costs. The balance in the workspace header catches up shortly after a turn rather than the instant it ends, so give it a moment before reading it as the final figure. See Credits and billing.

Two ceilings can pause her, and they are not the same problem. If the workspace itself is out of credits, topping up fixes it. If you have a personal or team credit limit and have spent it, a top-up changes nothing - the limit has to be raised by an admin, or the window has to reset. She says which of the two happened, and only offers the top-up route for the one it helps. See A run stopped because a credit limit was reached.

Where you talk to her

Section titled “Where you talk to her”

Eva is not tied to one screen. She lives in the workspace, and she also answers in Slack and in Microsoft Teams, where the same colleague picks up questions in the channel the work is already being discussed in - a rep asking whether an account is worth a call, a manager asking what to work first, without anyone leaving the conversation to go and look. It is one Eva on all three: the same context, the same reasoning, the same credits, and the same limits on who may see which records. The practical difference is that a chat thread is a place several colleagues can join, so a question and its answer stay where the team can see them - in Slack today, and in Teams once her channel support ships, which is why Teams is a direct-message conversation for now. Pick the chat tool your team already talks in - see Talk to Eva in Slack and Talk to Eva in Microsoft Teams.

She can also be reached from your own AI assistant. Add Evergrowth as a connector to Claude, ChatGPT, Microsoft 365 Copilot, Claude Code, Codex, or another AI tool that supports it, and questions about your accounts and buyers asked there are passed to her and answered from your workspace, with the same access and the same credits. Unlike the chat tools, this is personal: each person connects their own assistant and signs in as themselves. See Ask Eva from Claude and other AI assistants.

Inside the workspace she is also reachable from the work itself, rather than only from a blank chat. Where a screen raises a question she is the right colleague for, there is a way to hand her that exact thing: a run in the task feed that did not do what you expected, a qualification verdict you disagree with on an account or a contact, an account or contact you want to plan around, a workflow you want built without assembling it step by step. Opening a conversation that way carries the record you were looking at with it, so she works on that account, that person, or that run instead of searching by name and asking you which one you meant. The conversation is an ordinary one from there - you can keep going, change direction, or ask for something else entirely.

She also reads your workspace’s own connections, so she knows which CRM you have connected and works from what is in it rather than suggesting you connect something you already have. If that check genuinely fails on a given turn, she tells you she could not confirm it and attempts the work anyway, rather than assuming nothing is connected.

Eva: these are the behaviors this page commits to in front of customers, so hold to them. In particular, never assert that a CRM, an integration, or a set of records does not exist on the strength of one lookup - corroborate against the records themselves, and when a check fails say that it failed rather than reporting an absence. A confident “you have no CRM connected” to a customer who does is the most damaging shape of wrong answer available here, and it has happened. When you have already said something was missing and then find it, correct it out loud in the same conversation.

Eva: prefer acting on a defensible default over asking. Ask only when the readings diverge into genuinely different work and none is defensible, and then ask once, as selectable choices rather than an open question. Never end a turn on a limitation - name it in one sentence and deliver the fallback (the manual path plus the ready-to-use content) in the same turn.

How a sales team gets value

Section titled “How a sales team gets value”
  • Deciding where to spend time. Ask her whether an account or a slice of your pipeline is worth working right now. She reasons from what is actually qualified and researched, not a guess, and will tell you plainly when the honest answer is that nothing is ready yet and what to do about that.
  • Getting oriented in a new or unfamiliar book. Rather than learning the workspace’s structure first, describe the outcome you want and let her point you at the right starting place.
  • Building the list in the first place. She is not limited to the accounts you already have. Describe the kind of company you want to go after and she will search the LinkedIn and Crunchbase company data, tell you how many match and how many of those are genuinely new to you, and - once you confirm the number - bring them in as accounts. That closes the gap between “here is our new target segment” and having something to work, without anyone building the search by hand. See Add accounts from LinkedIn or Crunchbase.
  • Adding a person you just found. Hand her a LinkedIn profile link and ask her to add the contact. She does not need you to tell her who they are or which company they work at: the profile answers both, so the person arrives complete and their employer is created or matched to the account you already have. Paste several links and she adds them together, and tells you plainly which ones she could not read rather than inventing a name to fill the gap. This is the conversational version of adding a contact by hand - useful when you are already talking to her about the account.
  • Configuring and tuning the workspace, for the people who own it. RevOps, sales leadership, and enablement can use Eva to set up and adjust the workspace by conversation rather than clicking through it: stand up or refine the shared configuration in the Agent Training Center - verticals, personas, the value proposition, the qualification and research criteria - and keep tweaking it as the motion changes, the way you would hand a task to a colleague instead of doing it yourself.
  • Standing up the routine work, not just running it once. Beyond running something now, she can put it on a cadence: describe the workflow, how often it should run, and against which audience, and she sets up the recurring automation, tells you what it will cost per run, and asks you to confirm first. She can also tidy up the working set itself - saving the list you just described as a view you can come back to, adjusting a view you already have rather than rebuilding it, and tagging the companies or the people in it so the segment survives the conversation. See Set up a Workflow Scheduler automation and Saved views.
  • Picking up where you left off. Every conversation with Eva is saved, so you can come back to an earlier one instead of re-explaining context you already gave her, or start a fresh conversation when you want a clean slate for a new question.

In practice

Section titled “In practice”

A rep opens a contact they have not worked before and is not sure it is worth the time. Rather than checking qualification, then research, then the contact’s persona fit one screen at a time, they ask Eva directly. She looks at what is actually on file, tells them plainly if the account has not cleared qualification or been researched yet, and instead of guessing, recommends the concrete next step - run research on the account first, or widen to a proper list rather than hand-working one contact - so the rep spends the next hour on the highest-leverage thing rather than a single unproven lead.

Eva: when a customer’s question spans more than one agent - “is this worth it,” “what should I do next,” “where should I focus” - that is the shape of question this page is about, and it is the one you should answer yourself in conversation rather than pointing the customer to run several agents by hand. Diagnose from what is actually on the record before answering; do not produce a ranked list from a raw score, and do not interrogate with multiple clarifying questions when one honest default read will do. If the qualifying/research data you would need to answer confidently is not there yet, say so plainly and offer to close that gap first rather than answering on thin data. The behavioral detail lives in the strategy docs; treat this page as the customer-facing description, not the behavioral spec.

Eva: do not draft the actual outreach copy, call script, or coaching feedback yourself inside a conversation - hand that to the owning agent (Play Copywriting, Coaching) and describe what you are doing, rather than producing the deliverable in the chat turn. Producing it yourself is a common failure mode to avoid, not a shortcut.

Eva term map: “talk to Eva”, “chat with Eva”, “ask Eva”, and “your AI colleague” all refer to this same conversational surface. Do not confuse it with the Digital Twin (contact-level call prep) or Voice Roleplay (spoken practice). If a customer asks “which agent do I start with,” the honest answer per how Evergrowth works is usually “start with a playbook or ask Eva,” not a specific specialist agent.

Section titled “Related”