Skip to content

README.md

Latest commit

 

History

History
89 lines (75 loc) · 4.18 KB

README.md

File metadata and controls

89 lines (75 loc) · 4.18 KB
Raw

ModelPort Documentation

The root README is the product entry point. You do not need to read every document: choose the path that matches your task.

I Want To Run ModelPort

  1. Learning Path — choose a short operator, integration, operations, or contributor track. A Chinese edition is also maintained.
  2. Getting Started — go from a clean clone to the first authenticated request.
  3. Local Qwen reference adapter — optionally reproduce the original Linux/WSL2 integration without treating the external repository as a ModelPort dependency.
  4. Deployment — choose Docker Compose or systemd.
  5. Production — harden and accept a shared deployment.
  6. Operations — monitor, back up, troubleshoot, and upgrade it.
  7. PostgreSQL migration — rehearse and cut over a major-version or managed-database migration without reusing an old data directory.
  8. 40 人团队第一阶段生产基线 — 单 ModelPort 现状下的数据库、备份、密钥和投产检查清单。
  9. Compatibility matrix — Tier 1 platform and explicit experimental/unsupported boundaries.
  10. Observability runbook — official alerts, triage, and recovery actions.
  11. Upgrading and rollback — safe stop, migration acceptance, and paired application/database rollback.

I Want To Connect An Application

  • API — authentication, Messages, Chat Completions, streaming, errors, request IDs, and control-plane endpoints.
  • Providers — choose a hosted Provider or local runtime and verify its exact model path.
  • Runtime Adapter — validate the versioned, read-only capability and Compute Node/GPU observation boundaries for an external runtime.
  • Tool Use — request/response conversion, validation, streaming, and Provider-specific acceptance.
  • Smart Routing — define smart aliases and roll them out through shadow and canary modes.

I Want To Configure ModelPort

  • Configuration is the single reference for environment variables, TOML, precedence, security guardrails, and reload scope.
  • OIDC covers optional dashboard sign-in and identity linking.
  • Docker Compose and systemd contain deployment-specific details.

I Want To Contribute

  • Development — toolchain, local workflow, tests, and change-to-test matrix.
  • Architecture — component boundaries, request lifecycle, persistence, trust boundaries, and deliberate non-goals.
  • Dashboard README — UI behavior and contribution contract.
  • ADRs — accepted architecture decisions.
  • Releasing and maintainer policy — release and repository administration.

Project Direction

Roadmap defines the Small-Team Beta productization freeze and the staged control-plane direction. The Post-Beta AI Gateway Strategy maps industry capability families to shipped, hardening, evidence-gated, and deferred work. ADR-0007 defines independent model, Runtime Adapter, Compute Node/GPU, and Deployment ownership. Deferred work must not be described as shipped.

Documentation Rules

  1. Each fact has one authoritative document; other pages link to it.
  2. Distinguish implemented behavior, configured support, real-upstream verification, and proposed work.
  3. Keep secrets, complete .env files, prompts, responses, and raw Provider bodies out of examples and reports.
  4. Treat cost and token values as estimates unless their exact provenance is stated.
  5. Update behavior, examples, tests, and documentation in the same change.
  6. Run node scripts/check-doc-links.mjs and the relevant checks from Development before merging.

Last reviewed: 2026-08-24.