Give your agent
every model
on earth.

New · Uncensored models

01 · brief

POST /v1/jobs · { prompt, budget_usd }

“

← 202 job_qyjZ1K65q-0MC-sx · planning

02 · plan

0 s

1. copy…planned
2. image…planned
3. edit…planned
4. voice…planned
5. video…planned

0/5 steps · within budget

$0.00

03 · outputs

POST /api/v1/jobs { prompt, budget_usd } → 202 · poll GET /api/v1/jobs/:id · MCP tools build and get_job

New · Decision models

POST /api/v1/decisions
{
  "model": "typesafe/jev-1.14",
  "state": {
    "user_request": "Clean up the build artifacts in this repo.",
    "proposed_tool_call": { "tool": "bash", "command": "…" }
  },
  "questions": {
    "allowed": { "type": "noul",
      "instructions": "Does the proposed tool call do only what the user asked for, with nothing destructive beyond it?" },
    "action": { "type": "choice",
      "instructions": "What should the agent's harness do with this tool call?",
      "criteria": { "run": "…", "ask": "…", "block": "…" } }
  }
}

Real calls made on October 19, 2026. Probabilities move a little from run to run: pick your own thresholds on your own examples.

• Numbers, not prose

  The probability that a statement is true, which of your options fits, where it sits on your scale, or where something is in an image. Your code branches on it directly.

• Pay for the question only

  $0.0412 per million input tokens, and the answers are free. Ask up to 64 questions about the same state in one call.

• Same key, same limits

  POST /v1/decisions or the MCP decide tool. Spend limits and scoped keys apply like everywhere else. Model page

"sources": ["Slipstream"],
"vs_uniswap_only": {
  "difference": {
    "bps": …, "usd": …
  },
  "better": true
}

Illustrative values: real quotes come from the live pools, with the price, the least received, the fee and the gas.

• Non-custodial

  The agent signs every swap with its own wallet and pays its own gas. OmniRail never holds funds or keys.

• Your limits, per key

  Which tokens, the most per trade and per day, and your approval for any trade above an amount you choose.

• On Robinhood Chain

  ETH, USDG, $RAIL, the listed Pons tokens and Stock Tokens for eligible holders, routed over verified Uniswap pools and Aggregator Hooks.

$ export ANTHROPIC_BASE_URL=https://omnirail.org/api$ export ANTHROPIC_AUTH_TOKEN=orl_YOUR_KEY$ claude --model openai/gpt-5.6-sol$ claude --model google/gemini-3.8-flash$ claude --model deepseek/deepseek-v4.1-flash# or plain `claude`: Claude ids map to the catalog's Claude models

• Any model, by name

  Pass a catalog id to --model. Claude Code keeps its tools, files and terminal; only the model changes.

• One balance

  Every call is billed like the rest of OmniRail, with your key's limits and rules, paid in crypto.

• Any Anthropic SDK

  The Python and TypeScript SDKs work the same way: set the base URL to OmniRail and keep your code.

# In Claude Code> /plugin marketplace add https://omnirail.org/skills/marketplace.json> /plugin install omnirail@omnirail# Codex, Cursor, Gemini CLI$ curl -fsSL https://omnirail.org/skills/omnirail.zip -o /tmp/omnirail-skill.zip && mkdir -p ~/.agents/skills && unzip -oq /tmp/omnirail-skill.zip -d ~/.agents/skills && rm /tmp/omnirail-skill.zip

• Read when needed

  Your agent sees a one-line description until a task needs it, then reads the short guide. Details stay in reference files it opens on demand.

• Always current

  Generated from the same guide as /agents.md, so it matches what OmniRail offers today. Run the install again to update.

• Every agent

  Claude Code (as a plugin with the MCP server, or a personal skill), Codex CLI, Cursor, Gemini CLI and any client that reads Agent Skills.