Abel
Abel is prediction infrastructure for coding agents. It turns live numerical signals into Decision Tokens: structured prediction units with confidence, drivers, risks, and watch signals that agents can act on. Install Abel Skills in Codex, Claude Code, OpenCode, or OpenClaw with a single command, then ask your agent for forecasts grounded in causal computation instead of text completion.
The engine combines a time-series model that learns how signals evolve with a world causal graph refreshed daily across stocks, crypto, and macro data. Every prediction is measured against real outcomes, with public paper-traded portfolios and published research benchmarks backing the claims.
Abel targets developers building agent products, quantitative traders who need verifiable forecasts, and enterprise teams that want causal reasoning inside private deployments. The CAP protocol exposes predict, explain, intervene, and discover primitives through REST, MCP, and SDKs.
Install into Codex, Claude Code, OpenCode, or OpenClaw with one command
Decision Tokens ship 9 fields: prior, horizon, drivers, drift, watch signals, and action
Confidence never hits 1.0; every token closes with watch signals
CAP protocol exposes predict, explain, intervene, and discover over REST and MCP
Abel Portfolio publishes 110 live paper-traded strategies with public Sharpe metrics
6M-node causal graph refreshed daily across 200K+ variables
Structured Decision Tokens give agents verifiable predictions instead of open-ended text.
Free developer tier includes 1,000 API calls per month for Abel Skills.
Public paper-traded portfolios publish live Sharpe and return metrics for accountability.
CAP protocol supports REST, MCP, SDK, and direct-call integration modes.
No public self-serve pricing beyond the free developer tier; enterprise requires a sales demo.
Strongest causal graph coverage is in financial and macro domains with dense time-series data.
Confidence is intentionally capped below 1.0, so outputs always carry residual uncertainty.
Which coding agents support Abel Skills?
Abel Skills installs into Codex, Claude Code, OpenCode, and OpenClaw. Each platform has a one-line install instruction on abel.ai/skill that fetches the matching INSTALL.md from the Abel-skills GitHub repo.
What is a Decision Token in Abel?
A Decision Token is Abel's structured prediction output with nine fields: independent prior, time horizon, confidence, key drivers, anchor drift, watch signals, risk factors, and suggested action. Abel computes these from causal structure rather than generating text.
Does Abel have a free tier?
Yes. Abel offers a free developer tier with 1,000 API calls per month for Abel Skills users. Enterprise private deployment and BYOLLM options require contacting the Abel sales team.
How does Abel connect to agent workflows?
Abel integrates four ways: direct CAP API calls, schema-guided routing for LLMs, an MCP server for native tool use, and Abel Skills for zero-config agent onboarding. Docs live at cap.abel.ai.
What markets does Abel validate predictions against?
Abel validates predictions in financial markets with public paper-traded portfolios. Abel Equity Portfolio tracks 100 US stock members against QQQ, and Abel Crypto Portfolio tracks 10 crypto members against BTCUSD, with live metrics at abel.ai/abel-portfolio.
Can Abel run in a private enterprise deployment?
Yes. Abel Enterprise supports BYOLLM, private VPC deployment, custom causal graphs on proprietary data, SSO via SAML/OIDC, and three-zone graph access controls. Teams book a demo through abel.ai/enterprise.

