Hard Questions
What do I actually own if I cancel my AI agent tomorrow?
The inertia default
The inertia default running this question is the assumption that any vendor relationship deepens with use, and that by month six you will be too entangled to leave even if you wanted to. The intuition is reasonable. Most software relationships work that way — data accumulates, integrations multiply, team habits form, and the cost of leaving compounds with the cost of having stayed. The default tells you to be careful at signup or you will be trapped at month seven.
That instinct is correct in spirit. It is also too pessimistic about the specific case of an AI agent, because most of what compounds in an AI deployment lives in the team's hands, not in the vendor's database.
The slower thinking
The cancellation pattern for a Fidelic agent is the same pattern as any vertical AI agent. Sort what accumulates during the deployment by who actually has it; most of what is load-bearing is in your systems, not the vendor's, by virtue of where the agent worked.
The largest piece, by volume, is your Slack history. Every message the agent posted, every thread the team replied in, every brief and summary it attached, every escalation it flagged, every reasoning post explaining a refusal — all of it is in your Slack workspace, not the platform's. Slack retains those messages under your retention policy. Cancel the agent and the chat logs do not move. They are where they have been the whole time, in the channels the team reads.
Same logic for everything the agent wrote into a system that was already yours — drafts in your doc canvas, summaries attached to tickets in your support tool, entries in your CRM, updates to your wiki. The agent wrote into your systems in the first place; cancellation does not change that. The work product is in the same place a teammate's work would have been.
What is not yours by default is the agent's full activity log — the interior trace of how it perceived a situation, what it planned, what it tried, what it refused, the timestamps, the reasoning that did not make it into the public Slack post. That record lives on the platform. A buyer who wants it portable on cancellation arranges a chat-export add-on at deployment time. It is paid and has to be enabled before the agent ships, not after. Buyers who skip the add-on lose access to the interior record when they cancel. The public Slack posts remain because those are in your Slack; the deeper log does not.
Everything else stays with the platform, the way the underlying tooling stays with any vertical AI agent vendor. The constitution document, the configuration, the eval framework, the orchestration runtime, the Roster template the agent was instantiated from. The model weights are Anthropic's. None of that is yours to take, the same way you do not take Slack with you when you cancel Slack. The question is whether the things you do keep are the load-bearing ones. For most teams, the Slack history and the work product written into your existing systems are what was load-bearing the whole time.
The cancel-anytime guarantee on the pricing page commits to thirty days of notice for month-to-month buyers, with the option to lock in lower rates by committing to three or twelve months. If the deployment is wrong for you in week three, the cost of leaving is the price of one or two months. That is a small fraction of what most software lock-in patterns produce, and small enough that the decision to try the platform does not feel like a long marriage.
Sources
What would have to be true for the opposite to be correct
- Your Slack workspace does not retain messages under your team's retention policy, or you do not have admin access to your own Slack export
- The agent writes its outputs into vendor-only systems rather than into your existing Slack, docs, CRM, and ticketing
- You did not enable the chat-export add-on at deployment and you later need the agent's interior activity log
- The contract requires a multi-year commitment with no early-termination option
- Migration to a different vertical AI agent requires months of corpus rebuild and constitution rewriting that the team does not have time for
Where to next