liyanan b8980b68b8 fix(stream): stop silently swallowing upstream errors and session-lock stalls
Three stability fixes for the streaming layer. Together they eliminate
the "agent silently stops mid-task" failure mode users hit during agentic
loops.

1. Surface upstream errors instead of swallowing them

The SSE parsing loop wrapped every line in try { ... } catch {}, which
silently dropped the throw raised when Qoder returned an error envelope
(e.g. 406 Session blocked, provider_error). The stream then completed
with empty content and stopReason "stop", so pi saw a clean stop instead
of an error and never retried or informed the user.

The catch now only skips genuinely malformed SSE lines (SyntaxError) and
re-throws everything else so the outer handler can set stopReason "error"
+ errorMessage. A QODER_DEBUG env var logs skipped lines for diagnosis.

2. Stop overwriting the real finish_reason with "stop"

At end-of-stream the code unconditionally did:
    output.stopReason = toolCalls.length ? "toolUse" : "stop"
clobbering any meaningful finish_reason ("length", "content_filter", ...)
captured earlier. Now only toolUse is forced when tool calls are present;
otherwise the upstream finish_reason is preserved (defaults to "stop").
The done event reason cast now includes "length".

3. Give each request a fresh server-side session id

session_id was stableHash("qoder-session", userID, qoderModel) —
constant across all requests for a given user+model. Qoder enforces a
per-session lock and replies 406 Session blocked when a second request
arrives before the previous one fully drains server-side, which happens
routinely in tight agentic tool loops. Appending crypto.randomUUID()
makes every request use a fresh session. We send full conversation
history on each request, so losing server-side session continuity is
safe.

4. Capture upstream usage, responseId and responseModel

The provider reported all-zero usage and no responseId/responseModel.
Now the finish chunk usage (prompt/completion/total tokens, cache
read/write) is mapped into output.usage, and id/model into
responseId/responseModel, so pi sees real token accounting and can
correlate responses.

Add stream.test.ts with regression coverage for the 406 path, the
finish_reason preservation, the toolUse stop reason, and usage/response
metadata capture.
2026-07-25 20:55:40 +08:00
2026-07-20 18:22:35 +08:00

pi-provider-qoder

A pi provider extension that connects pi to the Qoder API, exposing Qoder Global and Qoder China models through provider surfaces.

Features

  • Two provider entries:
    • qoder — Global / international Qoder.
    • qoder-cn — Qoder China, forced to CN endpoints and independent of QODER_REGION.
  • Interactive Login: Global Qoder supports browser device-code flow or Personal Access Token (PAT) login.
  • Qoder CN PAT Login: China edition uses a separate PAT login entry (/login qoder-cn) and CN token exchange endpoints.
  • WAF Bypass: Built-in WAF obfuscation and body encoding (Encode=1).
  • COSY Signing: Full COSY signature header generation (RSA/AES-CBC/MD5).
  • Dynamic Model Catalog: Dynamically fetches model limits, effort configurations, and options from the /algo/api/v2/model/list endpoint.
  • Reasoning/Thinking Support: Real-time extraction of thinking process from API reasoning or HTML-like <think> tags.

Quick start

Install the provider:

pi install npm:pi-provider-qoder

Or install it globally with npm:

npm install -g pi-provider-qoder

Then log in from pi.

Global / international edition:

/login qoder

China edition:

/login qoder-cn

Personal Access Token (PAT)

A Qoder PAT (pt-...) cannot authenticate API calls directly — the provider exchanges it for a short-lived job token (mirroring the official qodercli / qoderclicn flow) and resolves your account identity automatically.

Global Qoder:

  • Run /login qoder and choose Use API Key (PAT), then paste the token.
  • Or set QODER_PERSONAL_ACCESS_TOKEN (or QODER_PAT) before starting pi.
  • QODER_API_KEY is also accepted; when set, pi automatically exchanges it and logs the provider in during startup.

Qoder China:

  • Run /login qoder-cn, then paste the CN PAT.
  • Or set QODERCN_PERSONAL_ACCESS_TOKEN (or QODERCN_PAT) before starting pi.
  • QODERCN_API_KEY is also accepted and triggers the same automatic startup login.

The exchanged job token is short-lived; the provider transparently re-exchanges the stored PAT when it expires.

Region environment variables

The provider also understands these optional variables:

export QODER_REGION=cn       # or QODER_BACKEND=cn / QODER_MODE=cn

Setting a CN PAT without a global PAT also auto-selects CN mode for the qoder entry, but the recommended explicit China entry is still /login qoder-cn and --provider qoder-cn.

Endpoints

Global:

  • PAT exchange: https://openapi.qoder.sh/api/v1/jobToken/exchange
  • User info: https://openapi.qoder.sh/api/v1/userinfo
  • Usage: https://openapi.qoder.sh/api/v2/quota/usage
  • Model / chat gateway: https://api3.qoder.sh/algo/api/v2/...

China:

  • PAT exchange: https://openapi.qoder.com.cn/api/v1/jobToken/exchange
  • User info: https://openapi.qoder.com.cn/api/v1/userinfo
  • Usage: https://openapi.qoder.com.cn/api/v2/quota/usage
  • Model / chat gateway: https://gateway.qoder.com.cn/algo/api/v2/...

Models

Global qoder

Exposes the backing model keys returned by Qoder, including:

  • Tier Models: auto, ultimate, performance, efficient, lite
  • Frontier Models:
    • qmodel (Qwen3.7 Plus)
    • qmodel_latest (Qwen3.7 Max)
    • dmodel (DeepSeek V4 Pro)
    • dfmodel (DeepSeek V4 Flash)
    • gm51model (GLM)
    • kmodel (Kimi)
    • mmodel (MiniMax)

China qoder-cn

The China provider exposes friendly model IDs and maps them back to Qoder CN's internal keys at request time:

Friendly ID Qoder CN key Context Images Reasoning
auto auto 180K ✅ ✅
qwen3.7-max qmodel_latest 1M ✅ ✅
qwen3.7-plus qmodel 1M ❌ ✅
qwen3.6-flash q36fmodel 1M ❌ ✅
deepseek-v4-pro dmodel 1M ❌ ✅
deepseek-v4-flash dfmodel 1M ❌ ❌
glm-5.2 gm51model 200K ✅ ✅
kimi-k2.6 kmodel 256K ✅ ✅
minimax-m2.7 mmodel 200K ❌ ❌

Compatibility aliases are also accepted for request mapping, such as qwen3.6-plus → qmodel, glm-5.1 → gm51model, and minimax-m3 → mmodel.

Usage

Once logged in, select any Qoder model in pi:

/model qwen3.7-plus

Or start directly:

pi --provider qoder-cn --model qwen3.7-plus

Global example:

pi --provider qoder --model auto

Architecture

src/
├── index.ts            # Extension registration
├── cosy.ts             # COSY signature, machine ID, region/endpoints, CN model aliases
├── login.ts            # OAuth device flow + PAT login sequence
├── pat.ts              # PAT → job-token exchange + identity resolution
├── models.ts           # Model definitions and dynamic config cache
├── oauth.ts            # PAT / OAuth callback orchestrator
├── stream.ts           # Main streaming response handler
├── transform.ts        # Message conversions (OpenAI schema mapping)
├── thinking-parser.ts  # Fallback <think> tag parser
└── qoder-encoding.ts   # WAF bypass body encoder

License

MIT

S
Description
No description provided
Readme
387 KiB
Languages
TypeScript 100%