[ Free ]
For putting the agent on your own site and seeing what it does with real visitors.
$0
100 credits / month
[ pricing ]
Start for free, no credit card required. Upgrade when you need a plan that fits your needs. No hidden fees.
[ Free ]
For putting the agent on your own site and seeing what it does with real visitors.
$0
100 credits / month
[ Business ]
For a live site with real traffic. Choose the credits you need and the price follows.
$50/month
[ Enterprise ]
For companies with several agents, a bigger team, and a procurement process.
Contact us
On every plan:
Everything in Free, plus:
Everything in Business, plus:
| Free | Business | Enterprise | |
|---|---|---|---|
| Usage limits | |||
| People | 1 | Up to 5 | As many as you need |
| Agents | 1 | 1 per workspace | Several per workspace |
| Workspaces | 1 | Several | Several |
| Credits a month | 100 | 2,000 to 200,000 | Negotiated |
| Most you can hold | 140 | 2,800 to 280,000 | 40% above the plan |
| Unused credits | Roll over, never expire | Roll over, never expire | Roll over, never expire |
| Free test replies a month | 200 | 1,000 | Negotiated |
| When credits run low | Pauses at zero, until the month turns | Auto-renew tops you up at your threshold, or renew early yourself | Negotiated |
| If nobody uses it | Deleted after 14 days without a conversation | Kept | Kept |
| The agent | |||
| Finds and connects to your site's own endpoints | ✓ | ✓ | ✓ |
| Co-navigates the page with the visitor | ✓ | ✓ | ✓ |
| Cards in chat | ✓ | ✓ | ✓ |
| Workflows and custom agents | ✓ | ✓ | ✓ |
| Books meetings on Google and Outlook calendars | ✓ | ✓ | ✓ |
| Enriches visitors through Apollo, LinkedIn and Exa | ✓ | ✓ | ✓ |
| Speaks the visitor's language | ✓ | ✓ | ✓ |
| Building it | |||
| Run it from Claude Code, Codex, Cursor, ChatGPT or Claude | ✓ | ✓ | ✓ |
| Stage changes, see the diff, keep the history | ✓ | ✓ | ✓ |
| Preview drafts on your live site | ✓ | ✓ | ✓ |
| A full trace of every workflow run | ✓ | ✓ | ✓ |
| A usage page that shows where every credit went | ✓ | ✓ | ✓ |
| Team and user management | – | ✓ | ✓ |
| Channels and handover | |||
| Web chat on your site | ✓ | ✓ | ✓ |
| Your team can take over any conversation live | ✓ | ✓ | ✓ |
| Handoff by email and Slack | ✓ | ✓ | ✓ |
| Systems behind a login, connected with our help | – | ✓ | ✓ |
| Escalation into your helpdesk, set up with our help | – | ✓ | ✓ |
| Controls | |||
| Domain allowlist for the widget | ✓ | ✓ | ✓ |
| A heads-up at 80% and again at 100% | ✓ | ✓ | ✓ |
| Volume rates, shown on your own usage page | – | – | ✓ |
| SSO, SLAs, a DPA and retention controls | – | – | ✓ |
| A security review and a named contact | – | – | ✓ |
| An annual contract, if procurement wants one | – | – | ✓ |
| Annual billing, with the year's credits up front | – | ✓ | ✓ |
| Feature | Type | Credits | Unit |
|---|---|---|---|
| Useful conversation* | A visitor the agent actually helped | 5 | / conversation |
| Conversation that went nowhere | Hellos, idle chats, dead ends | 0 | / conversation |
| Workflow or tool run | Whether the agent used it or a trigger fired it | 1 | / 10 runs |
| Test reply | Drafts, previews, your own test threads | 0 | / reply, up to your plan's allowance |
| Building over MCP | Any client, any change | 0 |
*A language model reads the closed conversation, along with what actually ran, and decides whether the agent helped. It runs our definition, not one the agent can be told to change. You see the verdict and its reason on every conversation, and you can contest it.
Bring your own provider keys and these cost nothing beyond the workflow run. LinkedIn is the exception: it always runs on ours. A cached Apollo result is never charged twice.
| Provider | Call | Credits | Unit |
|---|---|---|---|
| Apollo | Enrich a person or a company | 7 | / call |
| Apollo | Search people or companies | 2 | / call |
| Exa | Web search | 1 | / search |
| Glanos | Company lookup or search | 3 | / call |
| Enrich a profile or company; search people or companies | 3 | / call | |
| Posts, comments, insights, activity | 2 | / call |
This table only applies to custom agents and LLM workflow steps, where you wrote the prompt and picked the model. The agent's own replies to visitors are covered by the conversation price, whatever model we happen to run it on.
| Model | Provider | Credits | Unit |
|---|---|---|---|
| Claude Opus 4.7 | Anthropic | 300 | / 1M tokens |
| Claude Opus 4.6 | Anthropic | 300 | / 1M tokens |
| Claude Sonnet 4.6 | Anthropic | 180 | / 1M tokens |
| GPT-5.4 | OpenAI | 160 | / 1M tokens |
| Gemini 3.1 Pro | 130 | / 1M tokens | |
| GPT-5.3 | OpenAI | 120 | / 1M tokens |
| GPT-5.4 Mini | OpenAI | 50 | / 1M tokens |
| Gemini 3.1 Flash Lite | 15 | / 1M tokens | |
| GPT-5.4 Nano | OpenAI | 13 | / 1M tokens |
| Mistral Small 4 | Mistral | 8 | / 1M tokens |
The work the agent does, counted in credits. A conversation it helped with, a workflow it ran, a lookup it made, tokens a custom agent used: each draws a few credits from one balance, and your plan decides how many credits you get a month. There are no seats to buy and no fee when something converts.
Our unit of agent work. A useful conversation is five credits, a workflow run is a tenth of one, an enrichment lookup is somewhere between one and seven, and custom-agent tokens convert at a rate per model. The full pricing list above has every number.
Once per conversation, and only if it went somewhere. It does not matter how much the agent did in between: looked things up, ran a tool, booked a meeting, handed over to a person. When the conversation closes, a language model reads it alongside a record of what actually ran and decides whether the visitor was helped. A hello that goes nowhere costs nothing, and so does a dead end.
A language model, following our definition rather than anything the agent can be instructed to say. You see its verdict and its reasoning on every conversation, and you can contest one. If a visitor comes back to the same conversation because the answer did not hold, we reverse the charge. Your own test and preview traffic draws on a separate allowance of free test replies, 200 a month on Free and 1,000 on Business, and never touches your credits.
The usage page shows your balance and breaks it down by what drew it and by which agent or workflow did the drawing. So if one expensive custom agent is eating the budget, you will see that in a glance, not at the end of the month.
No. The amount you chose is the most you can spend in a month. A busy week burns credits faster, and what happens next is your choice: auto-renew bills the next month when your balance reaches the threshold you set, or you renew early by hand, which starts the month today and adds a full set of credits on top of what is left. Turn both off and the agent pauses at zero until the month turns. Either way the bill is a number you picked.
With auto-renew on, you do not get there: when your balance falls to the threshold you set, at most 40% of your plan, the next month is billed and its credits land on top of what is left. The two numbers are the same 40% on purpose, so a renewal at the threshold fills you exactly to your ceiling and nothing is wasted. With auto-renew off, we tell you at 80% and again at 100%, in the app and by email. At zero the agent pauses and the widget quietly comes off your site, rather than sitting there broken. You cannot buy extra credits, and there is never an overage invoice.
Yes, and they do not expire. What you do not spend stays with you from one month to the next, and from one year to the next. The limit is not time, it is size: your balance can hold up to 40% more than your plan's monthly credits, so a 2,000-credit plan tops out at 2,800. At the ceiling a renewal fills you to it rather than past it.
They come with you. An annual plan buys the year's credits up front, and the same rule sets the ceiling: 40% above the annual quota rather than above a month's. So whatever you had left on the monthly plan carries into a balance with a great deal more room in it, and a light last month before you switch costs you nothing.
Yes, and the credits you had left come with you either way. Raising it happens right away: you pay the new amount today and your month restarts today, on a bigger plan with a bigger ceiling. Lowering it applies from your next renewal. One thing to know before you go down: the ceiling is always 40% above whatever plan you are on now, so a smaller plan means a smaller buffer, and a balance above the new ceiling is trimmed to it. Nothing is prorated and nothing is taken away mid-month.
No. The agent does its own integration work. That is rather the point of it.
For room. Every plan gets the whole agent, including Free; Business is for a live site that needs more credits than a hundred, more than one person at the controls, and a workspace for each site you run. Enterprise is where the controls start: SSO, retention, a security review, and as many agents and people as you need.
One on Free, up to five on Business. If your team is bigger than that, talk to us about Enterprise, which is also where roles, SSO and retention controls live.
One site, one agent, one set of settings. Free gets one. Business lets you add a workspace for each site you run, each with its own agent. Enterprise can run several agents inside one workspace, for a site that needs a support agent and a sales agent side by side.
After 14 days without a single visitor conversation or agent run, it is deleted, agent and all. A widget on a live site keeps itself alive without you doing anything. Days spent paused at zero credits count as quiet, so if a Free account runs out early, the month needs to turn before two weeks pass. Paid plans are never deleted for being quiet.
No. Free and Business are self-serve, monthly, and you can cancel whenever you like. Enterprise has annual contracts for the companies that want them.
You keep your remaining credits until the end of the month you paid for. After that the account drops to Free.