Tolva · Agentic CRM
The agentic CRM
The agent does the reading, the thinking and the writing. You do the deciding.
Six AI agents research your leads, write your follow-ups and read your replies. Nothing reaches a customer, a deal or an inbox until you click Approve.
- Researches every new lead
- Drafts the follow-up email
- Reads replies, stops sequences
- Re-scores people who clicked
- Flags deals that went quiet
- Drafts proposals from the deal
Data consultancy, about 40 people, hiring two sales roles this quarter. Their site lists a new managed-analytics offer.
Founders
You sell and you build. The agents do the research and the first draft, so selling takes minutes a day.
Small sales teams
Two to ten people sharing one pipeline, one queue of proposals and one monthly AI budget.
B2B services
Consultancies and agencies whose deals start with a conversation and end with a proposal.
Growing software firms
Inbound and outbound in one place, with sequences that stop the moment someone answers.
How it works
Six agents, one funnel, your decision at every step
The approval gate
Every proposal waits for a person
Agents put their work on one review page. Start with the replies waiting, because those are people waiting on an answer. A brief has four outcomes, not two, so your answers teach you something.
- Approve. The research lands on the contact and the company, as a note on the timeline.
- Keep, no email. Keep the research and pause outreach to this person.
- Not a fit. The agent was right; this lead is not for you.
- Wrong. The agent got it wrong. This is the signal that prompts need tuning.
Built to be trusted
Four rules no agent can break
An automated check tests them against the product's source code, and any change that breaks one fails that check.
No agent can send
No agent has a tool that contacts a person. Agents cannot send email; that tool is absent, not switched off.
A person approves
A person approves every agent proposal before it changes a contact, a company, a deal or a campaign.
Every run is recorded
Every agent run is recorded with its tokens, its real web-search count, its duration and its cost, failures included.
The cap comes first
A monthly spend cap is checked before every run, not after.
AI costs you can see
Every run has a price, and you see it
A monthly cap is checked before every run, and failed runs count towards it, because a failed call is still billed. When the cap is reached the agents stop and say so.
The CRM underneath
Everything a small B2B team sells with
Sequences and campaigns
Multi-step email sequences with delays and open or click triggers, plus one-off campaigns, sent through SendGrid with open and click tracking.
- Steps keep their identity when you reorder or edit them, so nobody already enrolled is mailed twice.
- Bounces, spam reports and unsubscribes are suppressed automatically and stop every active sequence for that address.
- Every email carries an unsubscribe link and one-click List-Unsubscribe headers.
Contacts, pipeline and booking
Contacts with tags and full email history, companies that lead to their people, a drag-and-drop deal pipeline, rule-based lead scoring and a public booking page.
- Tags with autocomplete. Choosing two tags narrows the list to contacts that have both.
- A seven-stage pipeline is ready on the first day, with Kanban and list views.
- Your booking page offers only open slots in your timezone and books straight into the CRM.
- Your other apps can add contacts, and payments, through a token-protected ingest API that never creates a duplicate.
Your routine
Five minutes a day, and no agent acts while you are away
Daily
5 minutesOpen the review page. Answer the replies waiting first, then approve or reject the new briefs.
Weekly
20 minutesRe-score whoever is warm, ask for a rescue plan on the deals that went quiet, queue research for the week’s new contacts.
Monthly
one lookCheck the spend against the cap, and read the rejected pile: mostly “not a fit” means the channel, mostly “wrong” means the prompts.
Compare
Looking at a bigger CRM?
An honest comparison on what matters with agents: who sends, who approves, and what you can see of the cost.
Tolva vs HubSpot
HubSpot is a broad customer platform with marketing, sales, service and content hubs, and AI agents under the Breeze name. Tolva is a focused sales CRM for small B2B teams with six agents that only ever propose.
Tolva vs Salesforce
Salesforce is an enterprise CRM with agents under the Agentforce name, built for large organisations with administrators and integration teams. Tolva is ready on the first day for a small team, with no configuration project.
Questions people ask
What is an agentic CRM? +
An agentic CRM has AI agents inside it that read context, make a judgement and hand you the result. A classic CRM stores what you type. In Tolva, six agents research leads, draft follow-ups, re-score people who engaged, read replies, flag deals that went quiet and draft proposals. Each one ends by putting a proposal in front of a person.
Can Tolva's agents send emails on their own? +
No. No agent has a tool that sends email or contacts a person. That tool is absent, not switched off. An approved draft becomes a draft campaign, and sending it is a separate click by a person.
How much do the agents cost to run? +
Lead research costs USD 0.057 per contact, measured on real runs, because it searches the web. The other agents cost about USD 0.01 per run, and a proposal draft about USD 0.06. A hundred new contacts researched and drafted come to about USD 7. A monthly spend cap is checked before every run, and failed runs count towards it.
What happens when a prospect replies? +
Every active sequence for that person stops immediately, an explicit opt-out is honoured, and the reply is filed on their timeline. The reply is still forwarded to your normal inbox. The reply-reading agent then explains what the reply means and drafts an answer for you to approve.
See Tolva on a sample pipeline
In 30 minutes: a new contact researched, a follow-up drafted, a reply read and a proposal drafted, with every approval in between.
Request a demo