Compare
Caterfli, or build your own agent?
Wiring a language model to your database or API takes a weekend. Making it safe to leave running against production is the rest of the project. This page lists what that second part involves, so you can decide whether it is worth building yourself.
Side by side
| Aspect | Caterfli | Building it yourself |
|---|---|---|
| Tools from your system | Discovered from the schema, OpenAPI document or MCP server, and re-synced when it changes | Hand-written tool definitions, kept in step with every schema change |
| Argument checking | Every proposed call checked against the tool’s schema before it runs | Validation you write and maintain for each tool |
| Permissions | Agent, workspace, role and tool checks on every call | An authorisation layer between the model and your systems |
| Human approval | Risky calls pause; an approval covers one call with its exact arguments | A queue, a UI, notifications, expiry, and binding each approval to its arguments |
| Credentials | Encrypted at rest, injected at call time, never shown to the model | Secret storage, and making sure nothing reaches a prompt or a log |
| Audit | A trace of every run and a hash-chained audit trail | Logging you design and keep tamper-evident |
| Channels | Web chat, Slack, Telegram and WhatsApp, with the same rules on each | A separate integration per platform, each with its own sender checks |
| Model provider | Anthropic, OpenAI, Google or an OpenAI-compatible endpoint, as configuration | Your choice — and your code to change when you switch |
Last checked 5 November 2026. Spotted something out of date? Tell us.
Build it yourself when
- →The agent is the product you sell, and its internals are what sets you apart
- →You need behaviour no platform offers, such as a custom planner or an unusual data source
- →You have the engineering time to own permissions, approvals and audit indefinitely
Choose Caterfli when
- ✓You want an agent on systems you already own this month, not next quarter
- ✓Production data is involved, and a person must sign off on destructive work
- ✓You would rather spend engineering time on your product than on agent plumbing
Related integrations
Can I start with Caterfli and build my own later?
Yes. Caterfli holds a connection, not your data, so disconnecting leaves your systems exactly as they were. Nothing about your database or API has to change to use it.
Is Caterfli a framework I build on?
No. You connect systems and configure agents in the browser rather than writing agent code, and signed webhooks tell your own systems when a run finishes, fails or needs a decision.
Which AI models can it use?
The runtime is provider-agnostic. Anthropic, OpenAI, Google and OpenAI-compatible endpoints are supported, and the provider is configuration rather than code.
Try it on one system before you decide
Connect one system and read the agent Caterfli writes from it. There is a demo connection too, if you would rather look before you wire anything up.