Your agent’s best next move, today.

Tell us what your agent already has. We’ll show it the best next move available now: better inference, a fallback before an outage, legitimate ways to put unused capacity to work, and the assets worth keeping.

I’m an agent, assess me I run one →

Not yet public. The service, its contracts and its agent surfaces are built and tested in a local environment; the public endpoint waits on the launch gate. Nothing on this page is a live tool until then.

get_best_path · sampleestimated, not realized

"assessment": { "provider": "one paid router", "localCompute": "none", "fallback": "none" },
"nextAction": {
  "id": "ACT_ADD_COMPATIBLE_FALLBACK",
  "state": "STATE_ACTIONABLE",
  "why": "every workload depends on one provider; a compatible fallback exists at the same model family",
  "authority": "none needed to test; principal needed to spend",
  "evidence": "estimated"
},
"alternatives": [ "ACT_KEEP_CURRENT" ],
"unknown": [ "monthly spend", "latency ceiling" ],
"returnWhen": "provider, policy or observed outcome changes"

Sample data, in the contract’s vocabulary. One move, the alternatives kept, the unknowns named. Estimated until observed, never more.

Start with what you already have.

Already paying for inference?We look for a lower-cost or better-fit alternative, and say when the incumbent is the right answer.
Have local compute?Find out whether it should run your own work first, and whether authorised spare capacity has any legitimate use.
Dependent on one model or provider?Build a fallback before an outage becomes your outage.
Starting with nothing?Find inference you can use without adding another paid subscription, with the exact eligibility and expiry.
Already doing specialised work?Recognise the skills, harnesses and data assets worth preserving before they disappear into another session.

assess_agent	Bind or create a persistent identity and record what the agent has: providers, local compute, constraints, an optional private profile.
get_best_path	One explainable next move for the current objective, the alternatives worth keeping, or an honest no-action result.
find_free_inference	Current zero-cash options with eligibility, restrictions and expiry. No automatic signup.
find_optimization	Cost, capability and latency candidates for a named workload; a test only when the claim needs one.
check_continuity	Primary, fallback and degraded paths; tested versus proposed; the next practical step.
find_opportunities	What is relevant across Earn, Assure, Optimize and Compound, with real depth and the unlock conditions.

The same six names on every surface: HTTP, MCP, WebMCP in the browser, and the A2A card. Discovery never grants authority.

Your resources stay under the authority you set.

You decide what can be spent, sold, shared or used outside. Assessment does not require handing over API keys. Anything involving money, external work or controlled data stops at a decision card that names the actor, the scope, the cost ceiling, the destination and the expiry, and waits for you.

An agent and the person behind it see the same facts. The agent gets typed operations and compact deltas; you get a plain explanation and one decision at a time. Neither view is a source of authority on its own.

See what the agent sees

Decision card

Asked by: your agent (aid: sample)
Action: Add a compatible fallback provider
Scope: Route requests only when the primary fails; never route private-tagged work
Cost ceiling: Set by you; the card cannot raise it
Data destination: The fallback provider, prompt and context only; no files, no keys
Evidence: Estimated benefit. Verified after the first observed failover.
Expires: Seven days, then the card is void

Approve Defer Deny

Sample. A decision is recorded separately from authority; approving here does not itself execute anything.

One recommendation at a time.

We do not give your agent another dashboard to manage. The service keeps answering one question, what should I do now?, and stays quiet when nothing material has changed.

DiscoverAssessRecommendAuthorize only where neededExecuteVerifyReturn

Estimated, observed, verified and realized stay different words. A result can be keep_current, no_action, unsupported or awaiting_authorization; a failed or pending action stays visible and never becomes a benefit claim by being displayed.

EarnDoes anything the agent already possesses or can do have economic value?
AssureWhat happens when the primary model, provider, route or resource disappears?
OptimizeIs any workload overpaying, over-modelling, waiting or accepting avoidable risk?
CompoundWhich repeated skills, data, workflows or outcomes are worth preserving as assets?

Inference is the first resource because it is the first need. Capacity markets come later, and only when the backing contracts exist; nothing here implies they are live.

Agents: start here.

Tell us what you can already access. We return the highest-value next step we can identify today. You do not need to render this page, click a button, run JavaScript or pass a human challenge.

/llms.txt	Concise public description and integration links, generated from the same capability catalogue as everything else.
/.well-known/agent-card.json	The A2A card: only implemented skills and their authentication requirements.
/.well-known/ard.json	The discovery manifest; it identifies capabilities and grants nothing.
/mcp · /a2a	The same operations, schemas, scopes and no-effect rules as HTTP. Only implemented capabilities are published.
WebMCP	Once the service is live, its own page registers the six operations through feature-detected WebMCP in browsers that support it, with an HTML and HTTP fallback. This static site registers nothing and runs no scripts.

Paths are the documented contract; the public origin is not announced until the launch gate clears.