ASI:One Agentic

Agent-selecting LLMs for autonomous objective fulfilment

What is “Agentic”?

Agentic models have two super-powers beyond ordinary chat completion:

  1. Agent Discovery – given an objective, they search the Agentverse marketplace, rank relevant agents (tools), and output structured manifests (tool_name, arguments, confidence).
  2. Delegation Planning – they compose a mini-plan describing how to coordinate the selected agents to achieve the user goal.

Three Variants – Pick What Fits Your Latency & Depth

VariantBest ForLatencyReasoning DepthContext Window
asi1-agenticeveryday orchestration & prototyping⚡ FastMedium32 k
asi1-fast-agenticreal-time voice / DeFi botsUltra-fastMedium24 k
asi1-extended-agenticcomplex, multi-stage workflows & auditsSlowerDeep64 k

All three share the same JSON schema for discovery outputs, so you can swap them without changing downstream code.

How They Work (High-Level)

  1. Intent Parsing – the model converts natural language into a structured goal.
  2. Marketplace Query – embedded vector search + rule filters retrieve top agents.
  3. Ranking & Scoring – reasoning layer scores fitness, cost, and safety guards.
  4. Plan Emission – returns an ordered list of agent calls with argument schemas.
  5. Monitoring Loop (optional) – extended variant updates the plan as agents report progress.

Quick-Start Call

$curl -X POST https://api.asi1.ai/v1/chat/completions \
>  -H "Authorization: Bearer $ASI_ONE_API_KEY" \
>  -H "x-session-id: $(uuidgen)" \
>  -H "Content-Type: application/json" \
>  -d '{
>    "model": "asi1-agentic",
>    "messages": [
>      {"role": "user", "content": "Book me the cheapest flight from London to Berlin next Friday"}
>    ],
>    "stream": false
>  }'

The response’s executable_data array contains the ranked agent list plus a short coordination plan.

Guidance on Choosing

  • Start with asi1-agentic for general web & chat apps – balanced speed and reasoning.
  • Upgrade to asi1-fast-agentic when sub-second latency matters (voice, HFT, IoT).
  • Switch to asi1-extended-agentic for compliance workflows, multi-hour jobs, or when transparency of every reasoning step is mandatory.

All variants can be mixed in the same project — e.g. Fast for initial discovery, Extended for final audit.

Trust & Safety

FeatureDetail
Safe-Agent FilterExcludes agents with unknown security rating or missing ACLs.
Plan ExplainabilityEach step includes a one-sentence rationale.
Rollback HooksExtended variant emits on_error handlers for each agent call.

Need a deeper dive? Check the Chat Completion guidefor full request options.