SCHEMAXIS PILOT · Declared work. Results you can check.Documentation ↗ · Contact
schemaxis/Agent’s Hands

Delivery choices · offer under review

Core and optional hosting.

The five-point product explanation

  1. 01The agent declares. Schemaxis marshals the data.

    In the strict declared-work interface, the agent never runs code or directly handles source data. It declares where the data comes from or how it is generated, which operations it passes through, and how it is output. Schemaxis marshals immutable input captures, sealed execution and outputs retained by digest.

    This describes who operates on the inputs: no agent-supplied arbitrary scripts, shell or direct source-filesystem manipulation. The agent can receive bounded observations and receipt content; this does not promise that data is hidden from the model. Sealed container execution requires an explicitly configured, qualified backend. Native applications execute trusted runtime code. The historical local MCP demonstration alone does not establish that isolation.

  2. 02The plan is checked before anything executes.

    Admission, the declared schema, task scope and budget reservation are checked before execution. A request outside its authority is refused before a native job can run. Planning and inspection can run without launching the applications; native result verification is a separate step that reports the checks actually performed. In the retained MCP demonstration, a NumPy request under a SymPy-only scope returned SCOPE_EXCEEDED. Calls stopped at planning: zero native execution, all four prior NumPy output digests unchanged, and no successful current invocation to claim.

    Historical prepared-recipe local MCP demonstration, core 5cfe1b0. It is separate from both the comparative study and the hosted container-isolation proof.

  3. 03Smaller models failed alone, then produced passing artifacts.

    Haiku 4.5: 22/110 without → 110/110 with Schemaxis; Sonnet 4.6: 22/110 without → 110/110 with Schemaxis; GPT-5.6 Luna: 68/110 without → 110/110 with Schemaxis. Haiku 4.5’s native initial output had 32/32 images rejected, while its retained transcript called the result “production-ready”. The Schemaxis artifacts passed the required checks. The core supplies the verification state and receipt instead of relying on the agent’s success sentence. This records a misleading claim and changed artifact outcomes; it does not measure a general reduction in false-reporting propensity.

    29 September 2026: one authoring conversation per model and method, older model versions, 110 checks across an initial task and revision on small, purpose-built serial ParaView fixtures. Checks are not independent tasks. Some historical Schemaxis runs inspected implementation details or used helper code; one revision wrapper was broken despite passing artifacts. These are artifact outcomes, not qualification of the new strict no-shell authoring interface. The Opus native control was excluded after provider substitution.

  4. 04Reuse a supported recipe with varied parameters and a new receipt.

    A supported declared recipe can be repeated with varied authorised inputs and parameters; each execution produces its own receipt. The reusable document and bounded parameter contract are the mechanism, not a measured advantage for junior staff. Anyone using the supported, configured environment and authorised inputs can repeat it. Compatible runtimes, scope and checks still apply; the record preserves honest failure and unknown states.

    The downloadable two-case SymPy example checks local-core binding and host planning only: no native execution or new receipt was created by those preflight checks, and no hosted file-template endpoint is implied.

  5. 05Keep the frontier tradeoff and the cheaper-model result together.

    Frontier models passed either way in the retained paired studies, and Schemaxis was slower.

    GPT-5.6 Luna with Schemaxis passed all 110 checks at about one tenth of Astra’s native historical scored API cost (US$0.0815 versus US$0.8828); n=1 conversation each, older versions and 29 September 2026 prices. This is not current pricing or a total compute/support-cost estimate.

Recorded scope refusal ↗ Study artifacts and limits ↗

Use the core and Studio on your own infrastructure, or consider a provider-operated runtime. The earlier C$400 CAD per project / 14-day hosted proposal is under review. No final value proposition, price or current hosted access is established.

Free core + Studio

Apache-2.0 distribution boundary: declared operations, inspection/planning, reusable templates and parameter forms, local execution, verification and evidence. The source is available on GitHub; own compute, application rights, installation and maintenance are needed.

Workflow files

A template holds the recipe; parameter files describe particular jobs. Keeping those files does not keep a hosted account active. Local reuse needs fresh inputs, configuration and verification; old task/upload IDs and keys are not portable.

Optional hosting

Provider-operated execution, containment, fresh checks, result/receipt lifecycle, deletion controls and bounded support during access. It is not a charge for one conversion or unlimited use of all applications. Buyer value and the C$400 proposal remain under review.

No GUI paywall or licence change is adopted. A private team hub and integrated history/context service are separate directions, not included current runtime-pilot features. No paid subscription or maintained workflow library is being promised.

Earlier hosted proposal · under review

C$400 CAD

Per project · 14 elapsed days / 336 hours
Earlier draft before applicable tax; not a final offer.

  • The five admitted application scopes
  • One fixed fee, agreed in writing
  • No automatic renewal or usage charges
  • No automatic paid overflow
  • Your model-provider costs are separate
Prepare a local enquiry ↗

No purchase or payment action is available here.

Support from Schemaxis.

Weekdays 09:00–17:00 Winnipeg time, excluding locally observed holidays, with a next-business-day reply target. Two combined hours of onboarding and workflow support per project.

This is a reply target, not an uptime SLA. The monitored support channel still needs confirmation. If ordinary support would exceed two hours, pause and agree a narrower acceptance task or a refund decision. No extra charge is automatic.

Scope and capacity.

Proposed first cohort: up to two independent organisations and three projects. The final order must bind the release, task limits, data terms and absolute UTC start/end instants.

100 verified outcomes per UTC month is an upper quota, not a guarantee of success or a 100-run whole-trial entitlement. A month-spanning 14-day term can permit up to 200 across the two months; unresolved reservations reduce availability.

Input, operation and capacity bounds ↗

Prototype state explorer

What happens
at each stage?

Only preparation is current. Every later state below is a labelled design example. Changing this selector cannot activate, charge, run or delete anything.

All states remain documented without JavaScript in the expanded reference below.

Current real availability

Pilot in preparation

An enquiry can start a fit discussion. No account, payment or working customer access is available in this prototype.

No charge or credential.

Read all proposed states

Preparation / enquiry

An enquiry can start a fit discussion. No account, payment or working customer access is available in this prototype. No charge or credential.

Fit declined / waitlisted

The task is outside the admitted scope or capacity is not available. A fit discussion does not reserve a project. No charge or access promise.

Quote / terms pending

The earlier C$400 CAD / 14-day proposal is under review, not an issued offer. Price, scope, eligibility, tax, terms and exact UTC dates need the eventual reviewed order. Draft quote AH-PILOT-2026-01 · unsent.

Invoice unpaid / payment pending

An issued invoice, payment redirect, manually marked paid invoice or pending payout is not bank-reconciled settlement. No collection is connected here. No payment was requested or received.

Funds reconciled / access pending

In a real pilot, independently matched per-payment bank settlement would still be followed by host readiness, accepted terms and operator-confirmed activation. Synthetic example · no funds exist.

Active invited pilot

An approved project would use the invitation’s tested endpoint and short-lived key, with the final quota, support channel and absolute UTC expiry. No reusable key appears on the website. Synthetic example · no access has been issued.

Failed / unverified / unknown task

A failed or unverified result is not success. An unknown outcome requires operator reconciliation; do not replay it with a new idempotency key. No result or quota reset is implied.

Cancellation requested

Cancellation remains requested until the exact worker cleanup settles. Unknown outcomes and reservations stay visible. Requested is not completed.

Deletion pending

A deletion request is recorded, but worker reconciliation or physical removal is still pending. Poll the owned deletion status. Backup expiry remains separate.

Physical task deletion complete

Only actual deletion-status evidence could establish physical deletion. This example does not prove forensic erasure or expiry of every backup. Synthetic status · nothing was uploaded.

Expired / revoked

The 336-hour term ends at the agreed absolute UTC instant. No automatic renewal or silent extension. Arrange export, deletion and closure with support before expiry. Expiry does not certify backup deletion.

Refund pending / settled

A refund request or initiated refund is not a settled return. The agreed cancellation/refund terms and actual finance evidence govern each state. Draft policy not yet adopted; no payment or refund exists.

Incident / host loss

Suspend admissions, preserve unknown outcomes and contact support. Do not automatically replay jobs or clear reservations after restoring a host. No uptime SLA or recovery guarantee.

Before an invitation

The real release
still has prerequisites.

  • Business: customer eligibility, finance/tax owner, tax-inclusive total and adopted terms/refunds.
  • Operations: chosen host/region, final qualification, recovery evidence, secure credential delivery and monitored support.
  • Distribution: actual controlled domain/endpoint, licensing/source materials, accepted customer path and publication approval.
  • Collection: a verified collection method and bank reconciliation. Collection is not set up yet.
Inspect draft terms and data questions ↗

A small, deliberate pilot

Bring one repeatable task.
Agree what a good result means.

Enquire about Agent’s Hands