Jack Chen
cc2a83395a
DeepSeek retired the V4-Flash model and named the current generation `deepseek-flash`, with no version in the name. `deepseek-v4-flash` is still accepted, but DeepSeek documents it as a temporary compatibility name and already serves those requests from V4.1-Flash. The repository handed that name to users in the places most likely to be copied: the `oma` CLI default for `provider: 'deepseek'`, the adapter JSDoc that ships in the type declarations, the provider table in docs, the scaffolder env template, and every DeepSeek example. A copied name that the vendor calls temporary has an expiry date the reader cannot see. The provider table and the adapter JSDoc also stated that `deepseek-v4-flash` resolves to DeepSeek-V4-Flash-0731. That model is retired, so the sentence described a routing that no longer happens. Two corrections of the same kind travel with it. The thinking-mode note claimed the V4 models run at `high` effort, but the adapter forwards whatever effort the caller passes and sets none itself, so the claim is dropped rather than restated. Three `examples/basics` headers offered `deepseek-chat`, which DeepSeek retired outright on 2026-07-24 and which no longer resolves at all. The release bot moves to the same name. Callers that pass an explicit model are unaffected, and the old name keeps working for as long as DeepSeek accepts it. Left unchanged on purpose: `bench/config.json`, whose values are frozen controls for an already-run benchmark; the AtlasCloud pages, whose IDs come from that vendor's own model library; and the changelog history.
Express Customer Support
An Express REST API that wraps OMA's runTasks() explicit-DAG pipeline behind a single POST /tickets endpoint. A three-agent pipeline runs on every request:
- classifier (
deepseek-flash) — categorises the ticket and assigns urgency - drafter (
deepseek-v4-pro) — writes a polished customer-facing reply (depends on classifier) - qa-reviewer (
deepseek-v4-pro) — reviews the draft for tone and accuracy (depends on classifier + drafter)
runTasks() is the right primitive here because the pipeline shape is fixed: a coordinator-decomposed runTeam() would re-derive the same DAG and pay for an extra synthesis call on every request. With runTasks() the dependency graph is declared once at the route handler and OMA executes it in topological order, writing each agent's output into the next agent's prompt via the ## Context from prerequisite tasks block that runTasks() injects from each completed task's result.
Each agent uses a Zod outputSchema; the endpoint assembles the three structured results into one JSON response. Each agent's model and provider are independently swappable via env vars, so you can move any tier to another provider (see Swapping providers below).
Setup
cd packages/core/examples/integrations/express-customer-support
npm install
export DEEPSEEK_API_KEY=sk-... # default — all three agents use DeepSeek
Start the server
npm start
# Support API listening on http://localhost:3000
Send a ticket
curl -s -X POST http://localhost:3000/tickets \
-H 'Content-Type: application/json' \
-d '{"subject":"My order never arrived","body":"I placed order #12345 two weeks ago and it still has not shipped. Please help!"}' | jq .
Expected response shape:
{
"category": "shipping",
"urgency": "high",
"draft_reply": "Thank you for reaching out...",
"qa_notes": "Tone is empathetic and professional. Accuracy: ..."
}
Smoke test
Runs a real request against a local server (requires DEEPSEEK_API_KEY):
npm run smoke
Exits 0 and prints the structured response on success, exits 1 on failure.
HTTP error codes
| Status | Meaning |
|---|---|
| 200 | Pipeline completed; body matches the SupportTicketResponse schema |
| 400 | Invalid JSON body, or missing / non-string subject / body fields |
| 502 | Pipeline failed, an agent produced no structured output, or an LLM call threw |
| 504 | Pipeline exceeded the 60-second timeout (in-flight LLM calls are aborted) |
Environment variables
| Variable | Default | Description |
|---|---|---|
DEEPSEEK_API_KEY |
(required for default provider) | DeepSeek API key |
ANTHROPIC_API_KEY |
— | Required if anthropic is selected for any agent |
OPENAI_API_KEY |
— | Required if openai is selected for any agent |
GEMINI_API_KEY |
— | Required if gemini is selected for any agent |
XAI_API_KEY |
— | Required if grok is selected for any agent |
GITHUB_TOKEN |
— | Required if copilot is selected for any agent |
MINIMAX_API_KEY |
— | Required if minimax is selected for any agent |
AZURE_OPENAI_API_KEY |
— | Required if azure-openai is selected for any agent |
CLASSIFIER_PROVIDER |
deepseek |
Provider for the classifier agent |
CLASSIFIER_MODEL |
deepseek-flash |
Model for the classifier agent |
DRAFTER_PROVIDER |
deepseek |
Provider for the drafter agent |
DRAFTER_MODEL |
deepseek-v4-pro |
Model for the drafter agent |
QA_PROVIDER |
deepseek |
Provider for the QA reviewer agent |
QA_MODEL |
deepseek-v4-pro |
Model for the QA reviewer agent |
PORT |
3000 |
Port the server listens on |
Supported *_PROVIDER values with startup key validation: anthropic, openai, gemini, grok, deepseek, copilot, minimax, azure-openai. Providers not in this list are supported by the framework but are not mapped in this file's startup check — a missing key will surface as a 502 mid-request rather than a startup error.
Swapping providers
The defaults run the whole pipeline on DeepSeek — deepseek-flash for the cheap classifier, deepseek-v4-pro for the drafter and QA reviewer — so one DEEPSEEK_API_KEY runs everything. Each agent's provider and model are independently overridable via env vars, so you can move any tier to another provider.
Keep the cheap classifier on DeepSeek but run the drafter and QA reviewer on Anthropic:
export DEEPSEEK_API_KEY=...
export ANTHROPIC_API_KEY=sk-ant-...
export DRAFTER_PROVIDER=anthropic DRAFTER_MODEL=claude-sonnet-4-6
export QA_PROVIDER=anthropic QA_MODEL=claude-opus-4-8
npm start
Run the entire pipeline on OpenAI:
export OPENAI_API_KEY=sk-...
export CLASSIFIER_PROVIDER=openai CLASSIFIER_MODEL=gpt-4o-mini
export DRAFTER_PROVIDER=openai DRAFTER_MODEL=gpt-4o
export QA_PROVIDER=openai QA_MODEL=gpt-4o
npm start
If the API key for a chosen provider is missing, the server fails fast at startup with a clear message rather than erroring during a request.