A missing execution boundary for AI agents, trading bots, automations, and workflow systems.
When a system hits:
- timeout
- partial failure
- retry
- uncertain completion
…it often does not know whether the action already happened.
That is how you get:
- duplicate trades
- duplicate payments
- duplicate emails
- duplicate API mutations
SafeAgent adds a durable execution boundary around real side effects so retries can be reconciled instead of replayed.
Most systems can retry. Very few can decide when a retry is still correct.
A broker timeout does not mean the order failed.
It means the result is uncertain.
Without SafeAgent:
- the bot retries
- the same logical order is submitted again
- the position is duplicated
With SafeAgent:
- the retry resolves against the same execution record
- the result is reconciled
- the duplicate is blocked
This is not just a trading problem.
The same failure mode shows up in:
- payments
- browser workflows
- background jobs
- webhooks
- ticket creation
- agent tool calls
- external API mutations
pip install safeagent-exec-guardWithout SafeAgent: retry → replay → duplicate
With SafeAgent: retry → resolve → safe
- trading
- payments
- APIs
- agents
- workflows
Apache-2.0
