The model is the same.
In BootDesk, it is not alone.
OpenAI, Anthropic, and Google sell you the most capable models ever built, billed by the token. We use those same models under the hood, routed by task. The difference is everything around the model: the context it reads, the channels it answers on, the actions it takes through MCP tools and webhooks, the teammates it hands off to, and the records it leaves behind.
We are model-agnostic on purpose. BootDesk routes across providers by task — you get the better model when it lands, without reworking your setup.
The thesis
A model provider sells intelligence by the token. Intelligence is not a product. A product has context, hands, channels, teammates, and a record. BootDesk runs the same models and gives them all of the above. The moat is not the brain. The moat is the room the brain works in.
Feature by feature
A model endpoint, and a model with a job.
This is the one comparison where we do not claim a better model. We claim the same model, integrated. The table separates what the API gives you from what BootDesk gives the model.
| Capability | OpenAI · Anthropic · Google | BootDesk |
|---|---|---|
| What you get | A model endpoint, billed per token | The same models, wired into a full support workspace |
| Context the model sees | Whatever you paste into the prompt | The whole ticket history, the customer's memory, your systems via connectors |
| What the model returns | Text. You decide what to do with it | Drafts and actions, executed under your rules, in the ticket |
| Channels it answers on | Noneyou build the send and the threading | All of themreplies in the channel the customer used |
| Hands: tools & actions | You define and host every function call | MCP tools built in — GitHub, GitLab, Facebook — plus webhook-based tools for your own systems |
| Memory of the customer | None beyond what you pass each call | Persistent, opt-in, audited, across every ticket and channel |
| Handing off to a human | You build escalation, routing, and the context handover | Native. The assistant escalates with full context, on rules you set |
They build the models we run.
Here is when to go direct.
- 01Building your own product. If the model is the product, you want the API, full control of prompting, sampling, and safety, and the lowest raw token price.
- 02Fine-tuning and training. Custom weights, domain adaptation, eval pipelines. These belong at the provider layer.
- 03Frontier and research access. New modalities, long context research, early features. Direct first.
- 04Raw cost optimisation. At very large volume, your engineers tuning token usage will beat any bundled meter.
If your engineering team's job is the model, go to the source. We do.
- 01Context without copying. The model reads the whole ticket and history, not what an engineer remembered to paste.
If you have been building on a model API
Keep your prompts. Lose the plumbing.
You do not throw away model work to move to BootDesk. Your prompting, your guardrails, and your intent travel with you. What you drop is the orchestration, the threading, the send logic, and the maintenance.
Day one
Point BootDesk at your channels.
Point the assistant at the tickets your custom build was answering. It reads the same thread your prompts were trying to reconstruct from scratch.
Week one
Port your intents and rules.
Move your guardrails, escalation thresholds, and tone into BootDesk's rule layer. The model keeps doing what you taught it, now with action and handoff built in.
Month one
Retire the custom orchestration.
Decommission the glue code, the context assembly, the send pipelines. Your engineers move to your product. The model keeps working, inside the workspace.
How the bill stacks
The token, the orchestration,
and the engineer.
Direct model access is the cheapest raw intelligence you can buy. The cost is everything you build to make it useful in support: the context pipeline, the tool calls, the guardrails, the channels, the handoff, the logging. That work is invisible on the provider invoice and enormous on your payroll.
BootDesk folds the token cost and the integration into one consumption meter. Tokens are priced by tier on the public menu, every tool call is on the audit log, and the build disappears.
A direct model build
- Tokens in & out€220
- Vector store & retrieval€120
- Tool & function hosting€180
- Infra & monitoring€250
- Engineer time (amortised)€6,000
The token bill is small. The integration around it is the cost.
The same team, on BootDesk
- Commitment floorfrom €10
- 3 channels · 1.000 BDK each≈ €2.50
- 12 teammates · 3.000 BDK each≈ €30
- Assistant tokens (AI-heavy)≈ €60
- Context, retrieval, tools & channelsincluded
One meter. The model, the context, the hands, and the channels together.
Saves ~€6,665 / month
Incumbent side: illustrative composite. Direct token cost is genuinely lower per unit — the integration is what BootDesk removes. BootDesk side: real meter anatomy for an AI-heavy team.
The ones we hear from teams on the APIs.
If you use the same models, why not just call the API myself?
You can, and many teams should. The API gives you the model. What it does not give you is the context pipeline, the hands, the channels, the handoff, the audit, and the workspace around it. If your goal is support work, BootDesk is all of that, already assembled. If your goal is building a model-powered product, go direct.
Which model does BootDesk actually run?
Several, routed by task. Conversational replies use one tier, agentic work that reaches your systems uses another, longer creative drafts another. The choice is ours to update as providers improve — you get the better model when it lands, without reworking your setup.
Can the assistant reach our own systems?
Yes, two ways. Built-in MCP connectors cover GitHub, GitLab, and Facebook today. For everything else, webhook-based tools let you expose your own endpoints, and the assistant calls them like any other tool. Every call is visible in the thread and logged to the audit trail.
We have carefully tuned prompts. Do we lose them?
No. Your intents, guardrails, tone, and escalation thresholds move into BootDesk's rule layer. The model keeps behaving the way you taught it, with the bonus that it now reads the full thread, can act, and can hand off, instead of only returning text.
Is not direct API access cheaper?
Per token, yes, always. The model providers sell the cheapest intelligence on earth. The real cost of a support assistant is the integration around the model: the engineering to build it and maintain it. Once that is included, BootDesk tends to cost less overall, and the bill is easier to predict.
Can the assistant really act, or does it only draft?
Both, on rules you set. For low-risk, high-certainty work it acts directly: update a status, connect a conversation, call your webhook, close the loop in your tracker. For anything sensitive it drafts and hands to an agent for one keystroke. The threshold is yours, and every action is in the audit log.
Same model.
Given a job.
this page is an illustration, not a contract.