Praxa Execution Fabric

Current status: the production partner preview serves personal tenant keys, minimal durable-task execution, run reads, resumable SSE, cancellation requests, public and signed-in usage readback, automatic signed task webhooks, and bearer webhook management. The deployed portal accepts email-link/OTP fragment callbacks, persists the resulting browser session, and records legal acceptance durably before it renders developer routes. Exact recorded browser-action approval remains pending. Production canaries have separately proved zero-cost pre-dispatch cancellation and the scheduled non-retryable run_outcome_unknown terminal state for a genuinely provider-in-flight cancellation. An authenticated browser canary proved persisted platform-origin sessions, durable legal gating, and the canonical aliases; read-only HTTP checks proved nested developer routes and the consumer-host deep-link return. Turn execution, request-level overrides, organization execution and keys, and mature receipts remain closed. See Status and roadmap.

Source-ready follow-on: Portal-native personal agent/workflow management, a metered draft playground, profile/avatar/preferences, organization membership management, subscription display/eligible checkout, agent evaluation datasets, and a gated typed-workflow foundation now exist in source. They are not live claims until their migrations, Worker release, and production canaries are recorded on the status page. Experiments and model comparison remain unavailable.

​

Target approval flow

POST /v1/execute                submit a task, under a policy
        |
        v
   run starts                   routed across models and tools under that policy
        |
        v   (only if the policy calls for it)
   approval.required             event carries the exact action plus an approval digest
        |
        v
   you decide                   POST /v1/runs/:id/approve  { action_digest, decision }
        |
        v
   run finishes                 outcome + receipt: what ran, which policy, who approved, verified how

​

Status legend

​

Documentation

​

Getting started

• getting-started/overview.md — what the fabric is and how a request flows end to end
• getting-started/concepts.md — runs, events, receipts, policy, and tenancy in one page
• getting-started/quickstart.md — make your first authenticated call

​

API reference

• api/execute.md — POST /v1/execute: the minimal durable-task subset and reserved contract fields
• api/runs.md — GET /v1/runs/:id: read a run’s current state
• api/events.md — GET /v1/runs/:id/events: stream a run’s events, resume with Last-Event-ID
• api/approvals.md — exact digest-bound browser-action approval and fail-closed unsupported shapes
• api/usage.md — the usage contract and signed-in source implementation
• api/webhooks.md — active partner-preview signed delivery, ordering, and replay contract
• api/errors.md — the RFC 9457 error format every endpoint shares
• api/openapi.yaml — preview contract specification; Markdown status pages are activation authority

​

Keys & access

• keys-and-access/api-keys.md — creating, scoping, and rotating praxa_sk_ keys
• keys-and-access/scopes-and-tenancy.md — tenants, scopes, and the three principal types
• keys-and-access/rate-limits.md — current retry guidance and the unavailable dedicated per-key tiers

​

Approvals & policy

• approvals-and-policy/approvals.md — how a run pauses and how you resume it
• approvals-and-policy/policy-engine.md — current product gates and the planned unified engine
• approvals-and-policy/budgets-and-limits.md — budget ceilings and what happens when a run hits one

​

Studio & workflows

• studio-and-console.md — live, source-ready, gated, and unavailable developer-portal surfaces
• datasets.md — source-ready agent-scoped immutable evaluation inputs
• workflows.md — the gated typed-workflow source foundation and its activation boundary

​

Security

• security/security-model.md — tenant isolation and credential handling
• security/data-handling.md — current projections, unspecified retention, and unimplemented learning governance
• security/receipts-and-audit.md — current evidence versus planned receipts and exports

​

SDK

• sdk/typescript.md — the TypeScript client
• sdk/streaming.md — consuming a run’s event stream in code
• sdk/errors-and-retries.md — client-side retry and error-handling semantics

​

Use cases

• use-cases/overview.md — the three markets the fabric is built for
• use-cases/agent-builders.md — upgrading an existing custom agent onto the fabric
• use-cases/coding-and-desktop.md — cloud execution lanes for desktop coding sessions
• use-cases/business-workflows.md — internal org workflows on the same substrate

​

Enterprise

• enterprise/teams-and-organizations.md — org-level tenancy and billing
• enterprise/billing-and-plans.md — tiers, metering, and the plan ladder
• enterprise/deployment-and-isolation.md — dedicated isolation, and why there’s no self-hosted option

​

How-to guides

• how-to/build-an-approval-ui.md — render an approval payload correctly (render == record == what actually runs)
• how-to/receive-webhooks.md — verify signatures and handle retries
• how-to/bring-your-own-keys.md — connect your own provider credentials
• how-to/migrate-from-provider-apis.md — moving from a direct model API to the fabric

​

Reference

• faq.md — common questions, answered honestly
• status-and-roadmap.md — what’s shipped and what’s next, in customer terms
• glossary.md — terms used across these docs