Abel

Abel AI is a predictive service built for agents, and its technical base is causal computing. In practice that means it takes live inputs from whatever the agent is watching and converts them into decision tokens, which are compact outputs that tell the agent what its next move should be. The product also supplies independent priors, so an agent is not starting from nothing, and it calculates state transitions between steps, which is what makes the reasoning inspectable rather than opaque. The stated purpose is to give agents decision-making evidence that can be reasoned about instead of a pattern match that happens to look right. It installs on agent platforms including Codex, Claude Code, OpenCode, and OpenClaw, so it can be added to a setup that already exists. Pricing includes a free tier with 1,000 API calls per month.

- Turn live inputs into decision tokens that drive an agent's next action

- Use independent priors when there is little history to learn from

- Calculate state transitions between steps of a workflow

- Install the service into an existing agent platform such as Codex or Claude Code

- Test the whole approach on the free API tier before paying anything

- Developers building agents that have to choose actions on their own

- Teams already working inside Codex, Claude Code, OpenCode, or OpenClaw

- Engineers who want prediction logic they can audit rather than a black box

- Researchers and evaluators comparing causal approaches to plain prediction

The free tier is capped at 1,000 API calls per month, and both the cap and the rules around it can change, so production use needs a paid conversation with the vendor. Integration is real work rather than a switch you flip: someone has to wire the service into the agent loop and decide which inputs actually matter. Public documentation is thin, which raises the learning cost for a new team and means you should plan for experimentation rather than a smooth install. Finally, the output is evidence for a decision, not a guarantee that the decision is correct, and a causal model still depends on the quality of the inputs it receives.