# ujex

> ujex is open-source trust infrastructure for AI agents: identity, expiring authority, human approval, server-held secrets, and hash-chained evidence that reviewers can verify offline. It runs beneath an existing agent framework rather than replacing the agent loop.

Use this file as the curated map. Use the full reference for a single-context architecture, API, CLI, SDK, and workflow corpus. Product claims are deliberately bounded: evidence is tamper-evident, not tamper-proof against the operator; budget values are estimates; prompt-injection scoring is a risk signal; and ujex can only attest to actions that pass through its boundary.

The hosted proof loop is free and requires no card. The source tree, SDKs, CLI, MCP server, audit libraries, and deployment assets use Apache-2.0. Confirm the repository is public before linking it from a launch submission.

## Start here

- [Full LLM reference](https://ujex.dev/llms-full.txt): Single-file architecture, API, SDK, CLI, security, and usage reference.
- [Quickstart](https://docs.ujex.dev/getting-started/quickstart): Create an agent, run an approval-gated action, export evidence, and verify it offline.
- [Core concepts](https://docs.ujex.dev/getting-started/concepts): Agent, owner, credential, mandate, approval, evidence, secret, guardrail, and provenance vocabulary.
- [Build a first agent](https://docs.ujex.dev/getting-started/first-agent): TypeScript example for reading mail, requesting approval, replying, and writing provenance.

## Architecture and specifications

- [Architecture overview](https://docs.ujex.dev/architecture/overview): Control-plane topology, trust boundaries, request flow, custody limits, and deployment shape.
- [Evidence Pack format](https://docs.ujex.dev/): Append-only block format and offline verification model.
- [Evidence verification](https://docs.ujex.dev/compliance/evidence-verification): Passport, manifest, signature, trust-bundle, chain, and anchor verification.
- [Data model](https://docs.ujex.dev/architecture/data-model): Owner, agent, credential, mandate, approval, tool, secret, provenance, and audit records.
- [Reliability](https://docs.ujex.dev/architecture/reliability): Durable intent, retries, failure handling, scheduled controls, and offboarding.

## APIs and developer interfaces

- [OpenAPI document](https://ujex.dev/openapi.json): Machine-readable launch-critical HTTP and callable inventory.
- [API overview](https://docs.ujex.dev/api/overview): Authentication, callable envelopes, errors, idempotency, and API families.
- [Identity API](https://docs.ujex.dev/api/agent-identity): Sessions, credentials, mandates, delegation, secrets, and mailbox lifecycle.
- [Inbox API](https://docs.ujex.dev/api/agent-postbox): Agent mail, message reads, approval-gated sends, and signed bridge events.
- [Tools API](https://docs.ujex.dev/api/agent-tools): Server-side tool dispatch, policy checks, approval binding, and credential injection.
- [Provenance API](https://docs.ujex.dev/api/agent-memory): Markdown memory read, write, append, search, index, and delete operations.
- [SDK guide](https://docs.ujex.dev/getting-started/sdks): TypeScript, Python, Go, CLI, MCP, and framework integration paths.

## Trust and operational truth

- [Threat model](https://docs.ujex.dev/architecture/threat-model): Enforced boundaries, residual risks, and operator rules.
- [What ujex cannot prove](https://docs.ujex.dev/honest-limits/what-we-cannot-prove): Public limits on evidence, custody, credentials, spend, injection scoring, and assurance.
- [Security and disclosure](https://ujex.dev/security): Public trust material and vulnerability reporting.
- [Service status](https://ujex.dev/status): Hosted-surface status and monitoring limits.
- [Public verifier](https://ujex.dev/verify): Browser-local Evidence Pack verification.

## Optional

- [Homepage](https://ujex.dev/): Product positioning and trust-loop overview.
- [Pricing](https://ujex.dev/pricing): Current hosted access and packaging status.
- [Blog](https://ujex.dev/blog): Engineering explanations, comparisons, and operator guides.
- [Changelog](https://ujex.dev/changelog): Public product changes.
- [GitHub repository](https://github.com/axysar/ujex): Apache-2.0 source; publish this link only after repository visibility is confirmed.

<!-- global-site:start -->
## Global editions

- [EN homepage](https://ujex.dev/en): Ujex combines portable agent identity, scoped credentials, human approval, signed evidence, policy, secrets, metering, and an inbox behind SDKs and a native MCP server.
- [DE homepage](https://ujex.dev/de): Ujex verbindet portable Agentenidentität, begrenzte Zugangsdaten, menschliche Freigabe, signierte Nachweise, Richtlinien, Geheimnisse, Kostenkontrolle und Postfach über SDKs und einen nativen MCP-Server.
- [FR homepage](https://ujex.dev/fr): Ujex réunit identité portable, justificatifs limités, approbation humaine, preuves signées, politique, secrets, mesure et boîte de réception derrière des SDK et un serveur MCP natif.
- [ES homepage](https://ujex.dev/es): Ujex reúne identidad portátil, credenciales limitadas, aprobación humana, evidencia firmada, políticas, secretos, medición y buzón mediante SDK y un servidor MCP nativo.
- [JA homepage](https://ujex.dev/ja): Ujexは、移植可能なエージェントID、範囲付き資格情報、人の承認、署名済み証拠、ポリシー、秘密、計測、受信箱をSDKとネイティブMCPサーバーで提供します。
- [ZH homepage](https://ujex.dev/zh): Ujex通过SDK和原生MCP服务器提供可移植代理身份、范围凭据、人工批准、签名证据、策略、秘密、计量和收件箱。
- [SV homepage](https://ujex.dev/sv): Ujex samlar portabel agentidentitet, avgränsade behörigheter, mänskligt godkännande, signerade bevis, policy, hemligheter, mätning och inkorg bakom SDK:er och en inbyggd MCP-server.
- [HI homepage](https://ujex.dev/hi): Ujex portable agent identity, scoped credentials, human approval, signed evidence, policy, secrets, metering और inbox को SDK तथा native MCP server के पीछे एक साथ देता है।
- [Architecture](https://ujex.dev/en/architecture): Deep architecture, failure-boundary, and invariant reference.
- [Engineering hub](https://ujex.dev/engineering): Security, reliability, and trust-boundary engineering articles.
<!-- global-site:end -->