Update pi
This commit is contained in:
Martin Pander
@@ -40,56 +40,141 @@ export default function (pi: ExtensionAPI) {
|
||||
],
|
||||
});
|
||||
|
||||
// ── OpenAI via Langdock ─────────────────────────────────────────────────
|
||||
// ── OpenAI via Langdock (Responses API) ─────────────────────────────────
|
||||
// Chat Completions is unusable for agent work: Langdock rejects the combo pi
|
||||
// needs with "400 Function tools with reasoning_effort are not supported for
|
||||
// <model> in /v1/chat/completions. To use function tools, use /v1/responses
|
||||
// or set reasoning_effort to 'none'."
|
||||
// Docs: https://docs.langdock.com/en/developer/completion-api/openai-responses
|
||||
//
|
||||
// Verified against api.langdock.com/openai/eu/v1/responses:
|
||||
// ✓ SSE streaming, strict function tools, parallel calls, tool-result replay
|
||||
// ✓ image input, developer role, store:false (pi always sends it)
|
||||
// ✗ reasoning effort "minimal" (upstream rejects it for these models)
|
||||
// ✗ Lark/regex grammar tools → only `type: "function"` is allowed
|
||||
// ✗ max_output_tokens above the per-model cap (silently clamped)
|
||||
// • no encrypted reasoning is returned, but replaying reasoning items in
|
||||
// `input` on follow-up turns is accepted
|
||||
const openaiCompat = {
|
||||
// strict JSON-schema function tools are accepted
|
||||
supportsStrictMode: true,
|
||||
// Langdock: 'Unsupported tool type "custom". Only function tools are allowed.'
|
||||
supportsOpenAIGrammarTools: false,
|
||||
// skip the underscore-containing `session_id` header (proxies drop it),
|
||||
// keep x-client-request-id for session affinity
|
||||
sessionAffinityFormat: "openai-nosession",
|
||||
} as const;
|
||||
// NOTE: provider-level `compat` is ignored for extension-registered models,
|
||||
// so it has to be repeated on every model entry below.
|
||||
|
||||
// Langdock caps OpenAI models at 60k tokens/minute (Claude gets 200k), and a
|
||||
// request larger than the budget is rejected outright — so the usable window
|
||||
// is far below the model's real context size. 60k here makes pi auto-compact
|
||||
// at ~44k (contextWindow - reserveTokens) and stay under the gate.
|
||||
// Raise this if your workspace admin lifts the per-model TPM limit.
|
||||
const openaiContextWindow = 60000;
|
||||
|
||||
pi.registerProvider("openai", {
|
||||
baseUrl: "https://api.langdock.com/openai/eu/v1",
|
||||
apiKey: "$LANGDOCK_API_KEY",
|
||||
api: "openai-completions",
|
||||
api: "openai-responses",
|
||||
compat: openaiCompat,
|
||||
models: [
|
||||
{
|
||||
id: "gpt-5.6-sol",
|
||||
name: "GPT-5.6 Sol",
|
||||
reasoning: true,
|
||||
input: ["text", "image"],
|
||||
contextWindow: 272000,
|
||||
maxTokens: 128000,
|
||||
contextWindow: openaiContextWindow,
|
||||
maxTokens: 16000,
|
||||
thinkingLevelMap: {
|
||||
off: "none",
|
||||
minimal: null,
|
||||
low: "low",
|
||||
medium: "medium",
|
||||
high: "high",
|
||||
xhigh: "xhigh",
|
||||
max: "max",
|
||||
},
|
||||
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
|
||||
compat: openaiCompat,
|
||||
},
|
||||
{
|
||||
id: "gpt-5.6-terra",
|
||||
name: "GPT-5.6 Terra",
|
||||
reasoning: true,
|
||||
input: ["text", "image"],
|
||||
contextWindow: 272000,
|
||||
maxTokens: 128000,
|
||||
contextWindow: openaiContextWindow,
|
||||
maxTokens: 16000,
|
||||
thinkingLevelMap: {
|
||||
off: "none",
|
||||
minimal: null,
|
||||
low: "low",
|
||||
medium: "medium",
|
||||
high: "high",
|
||||
xhigh: "xhigh",
|
||||
max: "max",
|
||||
},
|
||||
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
|
||||
compat: openaiCompat,
|
||||
},
|
||||
{
|
||||
id: "gpt-5.6-luna",
|
||||
name: "GPT-5.6 Luna",
|
||||
reasoning: true,
|
||||
input: ["text", "image"],
|
||||
contextWindow: 272000,
|
||||
maxTokens: 128000,
|
||||
contextWindow: openaiContextWindow,
|
||||
maxTokens: 16000,
|
||||
thinkingLevelMap: {
|
||||
off: "none",
|
||||
minimal: null,
|
||||
low: "low",
|
||||
medium: "medium",
|
||||
high: "high",
|
||||
xhigh: "xhigh",
|
||||
max: "max",
|
||||
},
|
||||
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
|
||||
compat: openaiCompat,
|
||||
},
|
||||
{
|
||||
id: "gpt-5.5",
|
||||
name: "GPT-5.5",
|
||||
reasoning: true,
|
||||
input: ["text", "image"],
|
||||
contextWindow: 272000,
|
||||
maxTokens: 128000,
|
||||
contextWindow: openaiContextWindow,
|
||||
maxTokens: 16000,
|
||||
// "max" is rejected for this model
|
||||
thinkingLevelMap: {
|
||||
off: "none",
|
||||
minimal: null,
|
||||
low: "low",
|
||||
medium: "medium",
|
||||
high: "high",
|
||||
xhigh: "xhigh",
|
||||
max: null,
|
||||
},
|
||||
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
|
||||
compat: openaiCompat,
|
||||
},
|
||||
{
|
||||
id: "gpt-5.4-mini",
|
||||
name: "GPT-5.4 Mini",
|
||||
reasoning: false,
|
||||
input: ["text"],
|
||||
contextWindow: 272000,
|
||||
maxTokens: 16384,
|
||||
reasoning: true,
|
||||
input: ["text", "image"],
|
||||
contextWindow: openaiContextWindow,
|
||||
maxTokens: 32000,
|
||||
thinkingLevelMap: {
|
||||
off: "none",
|
||||
minimal: null,
|
||||
low: "low",
|
||||
medium: "medium",
|
||||
high: "high",
|
||||
xhigh: "xhigh",
|
||||
max: null,
|
||||
},
|
||||
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
|
||||
compat: openaiCompat,
|
||||
},
|
||||
],
|
||||
});
|
||||
|
||||
Reference in New Issue
Block a user