Most of NearSync costs nothing to run
A great deal of what feels like artificial intelligence in NearSync is not a language model at all. It is code that runs instantly, gives the same answer every time, and costs you nothing.
These are always free, on every plan, with no limit.
Quick capture. Type "call Sarah tomorrow at 3pm for 30 minutes p1 on teams" and NearSync reads the date, the time, the duration, the priority and the meeting platform straight out of the sentence. No model is involved. It is pattern matching, which is why it is instant and why it never gets creative with your calendar. The same goes for @team, @cockpit, @huddle and @remind, and for "on meet", "on teams" and "on zoom".
Only a genuinely compound instruction needs a model to untangle, something like "book the review at 4, then the client call straight after". Everything simpler stays free.
Automations and workflows. Every trigger, condition, branch, approval step, scheduled action and notification rule. A workflow is a rule you wrote, executed exactly as written. Building one costs nothing and running one costs nothing, however often it fires.
Everything you measure. Dashboards, charts, reports, metrics, forecasts and drill-downs.
Meeting analytics. Talk share, speaking turns, interruptions and acknowledgement rate are computed from the transcript arithmetically, not written by a model.
The pre-meeting brief, assembled from your own records before a call.
Finding things. Search, filters, saved views, duplicate detection and record matching.
Recorded meetings, up to 50 hours a month on Standard and 100 hours on Pro, shared across your whole organisation. That covers the recorder joining the call, the transcription with speakers identified, the summary and the coaching report. Meeting cost scales with how long your meetings run, and that is a poor thing to ask anyone to ration mid-conversation.
Email. Every notification, summary and automated message NearSync sends on your behalf, with no per-message charge and no volume tier.
What a credit is
A credit is NearSync's unit of AI usage. It is not a currency and it does not track any provider's bill.
One credit is about one question. Ask the assistant something and you have spent roughly one credit. Smaller jobs cost a fraction of that, so a single credit also covers around fifteen short summaries, or around twenty-five web chat replies.
Every seat comes with its own monthly allowance. It is yours rather than a shared workspace pot, so nothing you do consumes a colleague's capacity and nothing they do limits you. Unused credits do not roll over, and the allowance resets at the start of each billing month.
One thing does not come out of anyone's seat. When the web chat widget answers a visitor on your website, or WhatsApp routes an inbound message, nobody on your staff asked for it. Charging that to a seat would mean a busy week on your website quietly eating the capacity your sales team needs. That usage draws on a separate organisation allowance, available as an add-on. If you do not run web chat or WhatsApp intents, you do not need it.
What does spend credits
Credits are spent when you ask the assistant to write, read or listen to something.
The figures below are measured on OpenAI's GPT-5.4, which is the worked example throughout this page. Costs move with the model your organisation runs, and the section after next explains that.
Talking and listening
| Action | Roughly |
|---|---|
| A short voice conversation with the assistant | 25 credits |
| Dictating instead of typing | 2 to 3 credits a minute |
Speech is by a wide margin the most expensive thing you can spend credits on. A brief spoken exchange costs about as much as twenty-five typed questions. It is genuinely useful when your hands are busy, but it is worth knowing the difference.
This is talking to the assistant. Recording a meeting is free, within the hours above.
Asking and writing
| Action | Roughly |
|---|---|
| Audit knowledge base coverage | 6 credits |
| Recommend permissions for a role | 3 credits |
| Ask the assistant a question | 1 credit |
| Ask a question about your data | 1 credit |
| Draft or rewrite in a document | under 1 credit |
| Deal brief in the sales sidebar | about a tenth of a credit |
| Draft a reply to an email thread | about a tenth of a credit |
| Summarise or translate a message | about a fifteenth of a credit |
| Web chat answers a visitor | about a twenty-fifth of a credit |
A question that needs to look something up mid-answer costs two or three times a plain one, because the lookup is a second round trip.
Running in the background
| Action | Roughly |
|---|---|
| A scheduled agent run | under 2 credits |
| Indexing a document for search | negligible |
Agents are cheap individually and add up quietly, because they run whether or not anyone is watching.
Your model choice changes these numbers
Everything above is measured on GPT-5.4. Run something more capable, and each action costs proportionally more credits. Run something faster and lighter, and each costs less. The work is identical; only the engine changes.
Only an organisation administrator can change the model. Individual users cannot, which means your consumption stays predictable and one person cannot quietly move the whole workspace onto an expensive engine.
Inbound surfaces like web chat sit on an efficient model regardless. A visitor asking your opening hours does not need the most capable model available.
What makes one action cost more than another
Length of what it reads. A meeting transcript is thousands of times longer than a chat message. This dominates everything else.
How much context it needs. Answering "what should I do today?" means loading your work. Translating a sentence does not.
Whether it involves speech. Audio is expensive to process in both directions.
Keeping an eye on it
Your remaining allowance is shown as a percentage in the AI section of settings, covering your own seat and, if you administer it, the organisation allowance. There is little need to watch it day to day.
If an allowance runs out, work does not stop:
- Everything that is not AI keeps working exactly as before, which as the first section shows is most of the product
- Only the person who reached their limit is affected, because seats are separate
- The web chat widget hands the conversation to a person rather than going quiet
- You are told you have reached the limit rather than being handed a quietly worse answer
If you bring your own provider keys, none of this applies. You are billed by your provider directly and NearSync does not meter or cap your usage.
How these figures were derived
They come from a real month of usage: 1,145 requests, averaging 1,617 words of input and a short answer back. We rounded to numbers you can plan against rather than quoting decimals that would only ever apply to one example.
Your own consumption will vary with the length of your documents, the duration of your meetings and the model your organisation runs.
Did this answer your question?
No, ask a person