feat(ai): support PDF inputs (#38253)
This commit is contained in:
parent
8a36abd328
commit
2271f9b222
24 changed files with 1063 additions and 96 deletions
|
|
@ -235,9 +235,9 @@ describe("Anthropic Messages route", () => {
|
|||
}),
|
||||
)
|
||||
|
||||
// Regression: screenshot/read tool results must stay structured so base64
|
||||
// image data is not JSON-stringified into `tool_result.content`.
|
||||
it.effect("lowers image tool-result content as structured image blocks", () =>
|
||||
// Regression: read tool results must stay structured so base64 media data is
|
||||
// not JSON-stringified into `tool_result.content`.
|
||||
it.effect("lowers media tool-result content as structured blocks", () =>
|
||||
Effect.gen(function* () {
|
||||
const prepared = yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
|
||||
LLM.request({
|
||||
|
|
@ -253,6 +253,7 @@ describe("Anthropic Messages route", () => {
|
|||
result: [
|
||||
{ type: "text", text: "Image read successfully" },
|
||||
{ type: "file", uri: "data:image/png;base64,AAECAw==", mime: "image/png" },
|
||||
{ type: "file", uri: "data:application/pdf;base64,JVBERi0xLjQ=", mime: "application/pdf" },
|
||||
],
|
||||
}),
|
||||
],
|
||||
|
|
@ -263,6 +264,7 @@ describe("Anthropic Messages route", () => {
|
|||
expect(expectToolResult(prepared.body).content).toEqual([
|
||||
{ type: "text", text: "Image read successfully" },
|
||||
{ type: "image", source: { type: "base64", media_type: "image/png", data: "AAECAw==" } },
|
||||
{ type: "document", source: { type: "base64", media_type: "application/pdf", data: "JVBERi0xLjQ=" } },
|
||||
])
|
||||
}),
|
||||
)
|
||||
|
|
@ -292,7 +294,7 @@ describe("Anthropic Messages route", () => {
|
|||
}),
|
||||
)
|
||||
|
||||
it.effect("rejects non-image media in tool-result content with a clear error", () =>
|
||||
it.effect("rejects unsupported media in tool-result content with a clear error", () =>
|
||||
Effect.gen(function* () {
|
||||
const error = yield* LLMClient.prepare(
|
||||
LLM.request({
|
||||
|
|
@ -756,7 +758,7 @@ describe("Anthropic Messages route", () => {
|
|||
}),
|
||||
)
|
||||
|
||||
it.effect("continues a conversation with user image content", () =>
|
||||
it.effect("continues a conversation with user media content", () =>
|
||||
Effect.gen(function* () {
|
||||
const response = yield* LLMClient.generate(
|
||||
LLM.request({
|
||||
|
|
@ -766,6 +768,7 @@ describe("Anthropic Messages route", () => {
|
|||
Message.user([
|
||||
{ type: "text", text: "What is in this image?" },
|
||||
{ type: "media", mediaType: "image/png", data: "AAECAw==" },
|
||||
{ type: "media", mediaType: "application/pdf", data: "JVBERi0xLjQ=", filename: "report.pdf" },
|
||||
]),
|
||||
],
|
||||
}),
|
||||
|
|
@ -781,6 +784,7 @@ describe("Anthropic Messages route", () => {
|
|||
content: [
|
||||
{ type: "text", text: "What is in this image?" },
|
||||
{ type: "image", source: { type: "base64", media_type: "image/png", data: "AAECAw==" } },
|
||||
{ type: "document", source: { type: "base64", media_type: "application/pdf", data: "JVBERi0xLjQ=" } },
|
||||
],
|
||||
},
|
||||
],
|
||||
|
|
|
|||
|
|
@ -549,10 +549,12 @@ describe("Bedrock Converse route", () => {
|
|||
LLM.request({
|
||||
id: "req_doc",
|
||||
model,
|
||||
cache: "none",
|
||||
messages: [
|
||||
Message.user([
|
||||
{ type: "text", text: "Summarize these documents." },
|
||||
{ type: "media", mediaType: "application/pdf", data: "UERGREFUQQ==", filename: "report.pdf" },
|
||||
{ type: "media", mediaType: "text/csv", data: "Q1NWREFUQQ==" },
|
||||
{ type: "media", mediaType: "text/csv", data: "Q1NWREFUQQ==", filename: "data.csv" },
|
||||
]),
|
||||
],
|
||||
}),
|
||||
|
|
@ -563,10 +565,9 @@ describe("Bedrock Converse route", () => {
|
|||
{
|
||||
role: "user",
|
||||
content: [
|
||||
// Filename round-trips when supplied.
|
||||
{ text: "Summarize these documents." },
|
||||
{ document: { format: "pdf", name: "report.pdf", source: { bytes: "UERGREFUQQ==" } } },
|
||||
// Falls back to a stable placeholder when filename is missing.
|
||||
{ document: { format: "csv", name: "document.csv", source: { bytes: "Q1NWREFUQQ==" } } },
|
||||
{ document: { format: "csv", name: "data.csv", source: { bytes: "Q1NWREFUQQ==" } } },
|
||||
],
|
||||
},
|
||||
],
|
||||
|
|
@ -574,6 +575,96 @@ describe("Bedrock Converse route", () => {
|
|||
}),
|
||||
)
|
||||
|
||||
it.effect("requires names for document media", () =>
|
||||
Effect.gen(function* () {
|
||||
const error = yield* LLMClient.prepare(
|
||||
LLM.request({
|
||||
model,
|
||||
messages: [Message.user({ type: "media", mediaType: "application/pdf", data: "UERGREFUQQ==" })],
|
||||
}),
|
||||
).pipe(Effect.flip)
|
||||
|
||||
expect(error.message).toContain("document media requires a filename")
|
||||
}),
|
||||
)
|
||||
|
||||
it.effect("passes named document-only messages through for provider validation", () =>
|
||||
Effect.gen(function* () {
|
||||
const prepared = yield* LLMClient.prepare<BedrockConverse.BedrockConverseBody>(
|
||||
LLM.request({
|
||||
model,
|
||||
cache: "none",
|
||||
messages: [
|
||||
Message.user({
|
||||
type: "media",
|
||||
mediaType: "application/pdf",
|
||||
data: "UERGREFUQQ==",
|
||||
filename: "report.pdf",
|
||||
}),
|
||||
],
|
||||
}),
|
||||
)
|
||||
|
||||
expect(prepared.body.messages).toEqual([
|
||||
{
|
||||
role: "user",
|
||||
content: [{ document: { format: "pdf", name: "report.pdf", source: { bytes: "UERGREFUQQ==" } } }],
|
||||
},
|
||||
])
|
||||
}),
|
||||
)
|
||||
|
||||
it.effect("lowers document media in tool results", () =>
|
||||
Effect.gen(function* () {
|
||||
const prepared = yield* LLMClient.prepare<BedrockConverse.BedrockConverseBody>(
|
||||
LLM.request({
|
||||
model,
|
||||
messages: [
|
||||
Message.assistant([ToolCallPart.make({ id: "call_1", name: "read", input: { path: "report.pdf" } })]),
|
||||
Message.tool({
|
||||
id: "call_1",
|
||||
name: "read",
|
||||
result: {
|
||||
type: "content",
|
||||
value: [
|
||||
{ type: "text", text: "Read successfully" },
|
||||
{
|
||||
type: "file",
|
||||
uri: "data:application/pdf;base64,UERGREFUQQ==",
|
||||
mime: "application/pdf",
|
||||
name: "report",
|
||||
},
|
||||
],
|
||||
},
|
||||
}),
|
||||
],
|
||||
}),
|
||||
)
|
||||
|
||||
expect(prepared.body.messages).toEqual([
|
||||
{
|
||||
role: "assistant",
|
||||
content: [{ toolUse: { toolUseId: "call_1", name: "read", input: { path: "report.pdf" } } }],
|
||||
},
|
||||
{
|
||||
role: "user",
|
||||
content: [
|
||||
{
|
||||
toolResult: {
|
||||
toolUseId: "call_1",
|
||||
status: "success",
|
||||
content: [
|
||||
{ text: "Read successfully" },
|
||||
{ document: { format: "pdf", name: "report", source: { bytes: "UERGREFUQQ==" } } },
|
||||
],
|
||||
},
|
||||
},
|
||||
],
|
||||
},
|
||||
])
|
||||
}),
|
||||
)
|
||||
|
||||
it.effect("rejects unsupported image media types", () =>
|
||||
Effect.gen(function* () {
|
||||
const error = yield* LLMClient.prepare(
|
||||
|
|
|
|||
|
|
@ -70,6 +70,7 @@ describe("Gemini route", () => {
|
|||
Message.user([
|
||||
{ type: "text", text: "What is in this image?" },
|
||||
{ type: "media", mediaType: "image/png", data: "AAECAw==" },
|
||||
{ type: "media", mediaType: "application/pdf", data: "JVBERi0xLjQ=" },
|
||||
]),
|
||||
Message.assistant([ToolCallPart.make({ id: "call_1", name: "lookup", input: { query: "weather" } })]),
|
||||
Message.tool({ id: "call_1", name: "lookup", result: { forecast: "sunny" } }),
|
||||
|
|
@ -81,7 +82,11 @@ describe("Gemini route", () => {
|
|||
contents: [
|
||||
{
|
||||
role: "user",
|
||||
parts: [{ text: "What is in this image?" }, { inlineData: { mimeType: "image/png", data: "AAECAw==" } }],
|
||||
parts: [
|
||||
{ text: "What is in this image?" },
|
||||
{ inlineData: { mimeType: "image/png", data: "AAECAw==" } },
|
||||
{ inlineData: { mimeType: "application/pdf", data: "JVBERi0xLjQ=" } },
|
||||
],
|
||||
},
|
||||
{
|
||||
role: "model",
|
||||
|
|
@ -90,7 +95,12 @@ describe("Gemini route", () => {
|
|||
{
|
||||
role: "user",
|
||||
parts: [
|
||||
{ functionResponse: { name: "lookup", response: { name: "lookup", content: '{"forecast":"sunny"}' } } },
|
||||
{
|
||||
functionResponse: {
|
||||
name: "lookup",
|
||||
response: { name: "lookup", content: '{"forecast":"sunny"}' },
|
||||
},
|
||||
},
|
||||
],
|
||||
},
|
||||
],
|
||||
|
|
@ -110,7 +120,7 @@ describe("Gemini route", () => {
|
|||
}),
|
||||
)
|
||||
|
||||
it.effect("continues image tool results as inline vision input without base64 text", () =>
|
||||
it.effect("continues media tool results as inline model input without base64 text", () =>
|
||||
Effect.gen(function* () {
|
||||
const prepared = yield* LLMClient.prepare<Gemini.GeminiBody>(
|
||||
LLM.request({
|
||||
|
|
@ -125,6 +135,7 @@ describe("Gemini route", () => {
|
|||
value: [
|
||||
{ type: "text", text: "Image read successfully" },
|
||||
{ type: "file", uri: "data:image/png;base64,AAECAw==", mime: "image/png", name: "pixel.png" },
|
||||
{ type: "file", uri: "data:application/pdf;base64,JVBERi0xLjQ=", mime: "application/pdf" },
|
||||
],
|
||||
},
|
||||
}),
|
||||
|
|
@ -141,9 +152,12 @@ describe("Gemini route", () => {
|
|||
functionResponse: {
|
||||
name: "read",
|
||||
response: { name: "read", content: "Image read successfully" },
|
||||
parts: [
|
||||
{ inlineData: { mimeType: "image/png", data: "AAECAw==" } },
|
||||
{ inlineData: { mimeType: "application/pdf", data: "JVBERi0xLjQ=" } },
|
||||
],
|
||||
},
|
||||
},
|
||||
{ inlineData: { mimeType: "image/png", data: "AAECAw==" } },
|
||||
],
|
||||
},
|
||||
])
|
||||
|
|
@ -174,8 +188,13 @@ describe("Gemini route", () => {
|
|||
{
|
||||
role: "user",
|
||||
parts: [
|
||||
{ functionResponse: { name: "read", response: { name: "read", content: "" } } },
|
||||
{ inlineData: { mimeType: "image/jpeg", data: "/9j/" } },
|
||||
{
|
||||
functionResponse: {
|
||||
name: "read",
|
||||
response: { name: "read", content: "" },
|
||||
parts: [{ inlineData: { mimeType: "image/jpeg", data: "/9j/" } }],
|
||||
},
|
||||
},
|
||||
],
|
||||
},
|
||||
])
|
||||
|
|
@ -372,7 +391,10 @@ describe("Gemini route", () => {
|
|||
parts: [
|
||||
{ text: "thinking", thought: true },
|
||||
{ text: "", thought: true, thoughtSignature: "thought_sig" },
|
||||
{ functionCall: { name: "lookup", args: { query: "weather" } }, thoughtSignature: "tool_sig" },
|
||||
{
|
||||
functionCall: { id: "provider_call", name: "lookup", args: { query: "weather" } },
|
||||
thoughtSignature: "tool_sig",
|
||||
},
|
||||
],
|
||||
},
|
||||
finishReason: "STOP",
|
||||
|
|
@ -398,7 +420,10 @@ describe("Gemini route", () => {
|
|||
id: "reasoning-0",
|
||||
providerMetadata: { google: { thoughtSignature: "thought_sig" } },
|
||||
})
|
||||
expect(toolCall).toMatchObject({ providerMetadata: { google: { thoughtSignature: "tool_sig" } } })
|
||||
expect(toolCall).toMatchObject({
|
||||
id: "tool_0",
|
||||
providerMetadata: { google: { functionCallId: "provider_call", thoughtSignature: "tool_sig" } },
|
||||
})
|
||||
expect(response.events.findIndex((event) => event.type === "reasoning-end")).toBeLessThan(
|
||||
response.events.findIndex((event) => event.type === "tool-call"),
|
||||
)
|
||||
|
|
@ -416,6 +441,13 @@ describe("Gemini route", () => {
|
|||
providerMetadata: toolCall?.providerMetadata,
|
||||
}),
|
||||
]),
|
||||
Message.tool({
|
||||
id: "tool_0",
|
||||
name: "lookup",
|
||||
result: "done",
|
||||
resultType: "text",
|
||||
providerMetadata: toolCall?.providerMetadata,
|
||||
}),
|
||||
],
|
||||
}),
|
||||
)
|
||||
|
|
@ -424,7 +456,22 @@ describe("Gemini route", () => {
|
|||
role: "model",
|
||||
parts: [
|
||||
{ text: "thinking", thought: true, thoughtSignature: "thought_sig" },
|
||||
{ functionCall: { name: "lookup", args: { query: "weather" } }, thoughtSignature: "tool_sig" },
|
||||
{
|
||||
functionCall: { id: "provider_call", name: "lookup", args: { query: "weather" } },
|
||||
thoughtSignature: "tool_sig",
|
||||
},
|
||||
],
|
||||
},
|
||||
{
|
||||
role: "user",
|
||||
parts: [
|
||||
{
|
||||
functionResponse: {
|
||||
id: "provider_call",
|
||||
name: "lookup",
|
||||
response: { name: "lookup", content: "done" },
|
||||
},
|
||||
},
|
||||
],
|
||||
},
|
||||
])
|
||||
|
|
@ -498,7 +545,7 @@ describe("Gemini route", () => {
|
|||
content: {
|
||||
role: "model",
|
||||
parts: [
|
||||
{ functionCall: { name: "lookup", args: { query: "weather" } } },
|
||||
{ functionCall: { id: "tool_0", name: "lookup", args: { query: "weather" } } },
|
||||
{ functionCall: { name: "lookup", args: { query: "news" } } },
|
||||
],
|
||||
},
|
||||
|
|
@ -513,7 +560,13 @@ describe("Gemini route", () => {
|
|||
).pipe(Effect.provide(fixedResponse(body)))
|
||||
|
||||
expect(response.toolCalls).toEqual([
|
||||
{ type: "tool-call", id: "tool_0", name: "lookup", input: { query: "weather" } },
|
||||
{
|
||||
type: "tool-call",
|
||||
id: "tool_0",
|
||||
name: "lookup",
|
||||
input: { query: "weather" },
|
||||
providerMetadata: { google: { functionCallId: "tool_0" } },
|
||||
},
|
||||
{ type: "tool-call", id: "tool_1", name: "lookup", input: { query: "news" } },
|
||||
])
|
||||
expect(response.events.at(-1)).toMatchObject({ type: "finish", reason: "tool-calls" })
|
||||
|
|
|
|||
|
|
@ -5,6 +5,7 @@ import { LLM, LLMError, LLMEvent, Message, Model, ToolCallPart, ToolResultPart,
|
|||
import { Auth, LLMClient, RequestExecutor, WebSocketExecutor } from "../../src/route"
|
||||
import * as Azure from "../../src/providers/azure"
|
||||
import * as OpenAI from "../../src/providers/openai"
|
||||
import * as XAI from "../../src/providers/xai"
|
||||
import * as OpenAIResponses from "../../src/protocols/openai-responses"
|
||||
import * as ProviderShared from "../../src/protocols/shared"
|
||||
import { continuationRequest, nativeOpenAIResponsesContinuation } from "../continuation-scenarios"
|
||||
|
|
@ -16,6 +17,8 @@ const model = OpenAIResponses.route
|
|||
.with({ endpoint: { baseURL: "https://api.openai.test/v1/" }, auth: Auth.bearer("test") })
|
||||
.model({ id: "gpt-4.1-mini" })
|
||||
|
||||
const xaiModel = XAI.configure({ apiKey: "test", baseURL: "https://api.x.ai/v1" }).responses("grok-4.5")
|
||||
|
||||
const request = LLM.request({
|
||||
id: "req_1",
|
||||
model,
|
||||
|
|
@ -524,7 +527,77 @@ describe("OpenAI Responses route", () => {
|
|||
}),
|
||||
)
|
||||
|
||||
it.effect("rejects non-image media in tool-result content with a clear error", () =>
|
||||
it.effect("lowers PDF tool-result content as structured input_file array", () =>
|
||||
Effect.gen(function* () {
|
||||
const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
|
||||
LLM.request({
|
||||
id: "req_tool_result_pdf",
|
||||
model,
|
||||
messages: [
|
||||
Message.assistant([ToolCallPart.make({ id: "call_1", name: "read", input: {} })]),
|
||||
Message.tool({
|
||||
id: "call_1",
|
||||
name: "read",
|
||||
resultType: "content",
|
||||
result: [
|
||||
{
|
||||
type: "file",
|
||||
uri: "data:application/pdf;base64,JVBERi0xLjQ=",
|
||||
mime: "application/pdf",
|
||||
name: "report.pdf",
|
||||
},
|
||||
],
|
||||
}),
|
||||
],
|
||||
}),
|
||||
)
|
||||
|
||||
expect(expectToolOutput(prepared.body).output).toEqual([
|
||||
{
|
||||
type: "input_file",
|
||||
filename: "report.pdf",
|
||||
file_data: "data:application/pdf;base64,JVBERi0xLjQ=",
|
||||
},
|
||||
])
|
||||
}),
|
||||
)
|
||||
|
||||
it.effect("uses xAI inline file encoding for PDF tool results", () =>
|
||||
Effect.gen(function* () {
|
||||
const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
|
||||
LLM.request({
|
||||
model: xaiModel,
|
||||
messages: [
|
||||
Message.assistant([ToolCallPart.make({ id: "call_1", name: "read", input: {} })]),
|
||||
Message.tool({
|
||||
id: "call_1",
|
||||
name: "read",
|
||||
resultType: "content",
|
||||
result: [
|
||||
{
|
||||
type: "file",
|
||||
uri: "data:application/pdf;base64,JVBERi0xLjQ=",
|
||||
mime: "application/pdf",
|
||||
name: "report.pdf",
|
||||
},
|
||||
],
|
||||
}),
|
||||
],
|
||||
}),
|
||||
)
|
||||
|
||||
expect(expectToolOutput(prepared.body).output).toEqual([
|
||||
{
|
||||
type: "input_file",
|
||||
filename: "report.pdf",
|
||||
file_data: "JVBERi0xLjQ=",
|
||||
mime_type: "application/pdf",
|
||||
},
|
||||
])
|
||||
}),
|
||||
)
|
||||
|
||||
it.effect("rejects unsupported media in tool-result content with a clear error", () =>
|
||||
Effect.gen(function* () {
|
||||
const error = yield* LLMClient.prepare(
|
||||
LLM.request({
|
||||
|
|
@ -1526,20 +1599,64 @@ describe("OpenAI Responses route", () => {
|
|||
}),
|
||||
)
|
||||
|
||||
it.effect("lowers user image content", () =>
|
||||
it.effect("lowers user image and PDF content", () =>
|
||||
Effect.gen(function* () {
|
||||
const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
|
||||
LLM.request({
|
||||
id: "req_media",
|
||||
model,
|
||||
messages: [Message.user({ type: "media", mediaType: "image/png", data: "AAECAw==" })],
|
||||
messages: [
|
||||
Message.user([
|
||||
{ type: "media", mediaType: "image/png", data: "AAECAw==" },
|
||||
{ type: "media", mediaType: "application/pdf", data: "JVBERi0xLjQ=", filename: "report.pdf" },
|
||||
]),
|
||||
],
|
||||
}),
|
||||
)
|
||||
|
||||
expect(prepared.body.input).toEqual([
|
||||
{
|
||||
role: "user",
|
||||
content: [{ type: "input_image", image_url: "data:image/png;base64,AAECAw==" }],
|
||||
content: [
|
||||
{ type: "input_image", image_url: "data:image/png;base64,AAECAw==" },
|
||||
{
|
||||
type: "input_file",
|
||||
filename: "report.pdf",
|
||||
file_data: "data:application/pdf;base64,JVBERi0xLjQ=",
|
||||
},
|
||||
],
|
||||
},
|
||||
])
|
||||
}),
|
||||
)
|
||||
|
||||
it.effect("uses xAI inline file encoding for user PDFs", () =>
|
||||
Effect.gen(function* () {
|
||||
const prepared = yield* LLMClient.prepare<OpenAIResponses.OpenAIResponsesBody>(
|
||||
LLM.request({
|
||||
model: xaiModel,
|
||||
messages: [
|
||||
Message.user({
|
||||
type: "media",
|
||||
mediaType: "application/pdf",
|
||||
data: "data:application/pdf;base64,JVBERi0xLjQ=",
|
||||
filename: "report.pdf",
|
||||
}),
|
||||
],
|
||||
}),
|
||||
)
|
||||
|
||||
expect(prepared.body.input).toEqual([
|
||||
{
|
||||
role: "user",
|
||||
content: [
|
||||
{
|
||||
type: "input_file",
|
||||
filename: "report.pdf",
|
||||
file_data: "JVBERi0xLjQ=",
|
||||
mime_type: "application/pdf",
|
||||
},
|
||||
],
|
||||
},
|
||||
])
|
||||
}),
|
||||
|
|
@ -1551,11 +1668,11 @@ describe("OpenAI Responses route", () => {
|
|||
LLM.request({
|
||||
id: "req_media",
|
||||
model,
|
||||
messages: [Message.user({ type: "media", mediaType: "application/pdf", data: "AAECAw==" })],
|
||||
messages: [Message.user({ type: "media", mediaType: "application/x-tar", data: "AAECAw==" })],
|
||||
}),
|
||||
).pipe(Effect.flip)
|
||||
|
||||
expect(error.message).toContain("OpenAI Responses does not support media type application/pdf")
|
||||
expect(error.message).toContain("OpenAI Responses does not support media type application/x-tar")
|
||||
}),
|
||||
)
|
||||
|
||||
|
|
|
|||
207
packages/ai/test/provider/pdf.recorded.test.ts
Normal file
207
packages/ai/test/provider/pdf.recorded.test.ts
Normal file
|
|
@ -0,0 +1,207 @@
|
|||
import { describe, expect } from "bun:test"
|
||||
import { Effect, Schema, Stream } from "effect"
|
||||
import { LLM, LLMResponse, Message, ToolDefinition, type Model } from "../../src"
|
||||
import { AmazonBedrock, Anthropic, Google, OpenAI, XAI } from "../../src/providers"
|
||||
import { LLMClient } from "../../src/route"
|
||||
import { Tool } from "../../src/tool"
|
||||
import { runTools } from "../lib/tool-runtime"
|
||||
import { recordedTests } from "../recorded-test"
|
||||
|
||||
const CODE = "ORCHID-7391"
|
||||
const PDF =
|
||||
"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"
|
||||
|
||||
const openai = OpenAI.configure({ apiKey: process.env.OPENAI_API_KEY ?? "fixture" })
|
||||
const anthropic = Anthropic.configure({ apiKey: process.env.ANTHROPIC_API_KEY ?? "fixture" })
|
||||
const google = Google.configure({ apiKey: process.env.GOOGLE_API_KEY ?? "fixture" })
|
||||
const xai = XAI.configure({ apiKey: process.env.XAI_API_KEY ?? "fixture" })
|
||||
const bedrock = AmazonBedrock.configure({
|
||||
apiKey: process.env.AWS_BEDROCK_API_KEY ?? "fixture",
|
||||
region: process.env.AWS_REGION ?? "us-east-1",
|
||||
})
|
||||
|
||||
const targets: ReadonlyArray<{
|
||||
readonly id: string
|
||||
readonly name: string
|
||||
readonly provider: string
|
||||
readonly protocol: string
|
||||
readonly requires: string
|
||||
readonly filename: string
|
||||
readonly maxTokens: number
|
||||
readonly model: Model
|
||||
}> = [
|
||||
{
|
||||
id: "openai",
|
||||
name: "OpenAI Responses gpt-4o-mini",
|
||||
provider: "openai",
|
||||
protocol: "openai-responses",
|
||||
requires: "OPENAI_API_KEY",
|
||||
filename: "verification.pdf",
|
||||
maxTokens: 40,
|
||||
model: openai.responses("gpt-4o-mini"),
|
||||
},
|
||||
{
|
||||
id: "anthropic",
|
||||
name: "Anthropic Haiku 4.5",
|
||||
provider: "anthropic",
|
||||
protocol: "anthropic-messages",
|
||||
requires: "ANTHROPIC_API_KEY",
|
||||
filename: "verification.pdf",
|
||||
maxTokens: 40,
|
||||
model: anthropic.model("claude-haiku-4-5-20251001"),
|
||||
},
|
||||
{
|
||||
id: "gemini",
|
||||
name: "Gemini 3.5 Flash",
|
||||
provider: "google",
|
||||
protocol: "gemini",
|
||||
requires: "GOOGLE_API_KEY",
|
||||
filename: "verification.pdf",
|
||||
maxTokens: 256,
|
||||
model: google.model("gemini-3.5-flash"),
|
||||
},
|
||||
{
|
||||
id: "xai",
|
||||
name: "xAI Grok 4.5",
|
||||
provider: "xai",
|
||||
protocol: "openai-responses",
|
||||
requires: "XAI_API_KEY",
|
||||
filename: "verification.pdf",
|
||||
maxTokens: 40,
|
||||
model: xai.responses("grok-4.5"),
|
||||
},
|
||||
{
|
||||
id: "bedrock",
|
||||
name: "Bedrock Claude Haiku 4.5",
|
||||
provider: "amazon-bedrock",
|
||||
protocol: "bedrock-converse",
|
||||
requires: "AWS_BEDROCK_API_KEY",
|
||||
filename: "verification",
|
||||
maxTokens: 40,
|
||||
model: bedrock.model("us.anthropic.claude-haiku-4-5-20251001-v1:0"),
|
||||
},
|
||||
]
|
||||
|
||||
const recorded = recordedTests({ prefix: "pdf", tags: ["pdf"] })
|
||||
const prompt = "Return only the verification code from the PDF."
|
||||
const readPdf = ToolDefinition.make({
|
||||
name: "read_pdf",
|
||||
description: "Read the attached PDF.",
|
||||
inputSchema: { type: "object", properties: {}, additionalProperties: false },
|
||||
})
|
||||
const readPdfRuntime = Tool.make({
|
||||
description: readPdf.description,
|
||||
parameters: Schema.Struct({ path: Schema.String }),
|
||||
success: Schema.String,
|
||||
execute: () => Effect.succeed("PDF read successfully"),
|
||||
toModelOutput: () => [
|
||||
{ type: "text", text: "PDF read successfully" },
|
||||
{
|
||||
type: "file",
|
||||
uri: `data:application/pdf;base64,${PDF}`,
|
||||
mime: "application/pdf",
|
||||
name: "verification.pdf",
|
||||
},
|
||||
],
|
||||
})
|
||||
|
||||
const expectCode = (response: LLMResponse) => {
|
||||
expect(response.finishReason).toBe("stop")
|
||||
expect(response.text.toUpperCase()).toContain(CODE)
|
||||
}
|
||||
|
||||
describe("PDF recorded", () => {
|
||||
for (const target of targets) {
|
||||
recorded.effect.with(
|
||||
`reads a user PDF with ${target.name}`,
|
||||
{
|
||||
id: `${target.id}-user-input`,
|
||||
provider: target.provider,
|
||||
protocol: target.protocol,
|
||||
requires: [target.requires],
|
||||
tags: ["user-input"],
|
||||
},
|
||||
Effect.gen(function* () {
|
||||
expectCode(
|
||||
yield* LLMClient.generate(
|
||||
LLM.request({
|
||||
id: `recorded_pdf_${target.id}_user_input`,
|
||||
model: target.model,
|
||||
cache: "none",
|
||||
generation: { maxTokens: target.maxTokens, temperature: 0 },
|
||||
messages: [
|
||||
Message.user([
|
||||
{ type: "media", mediaType: "application/pdf", data: PDF, filename: target.filename },
|
||||
{ type: "text", text: prompt },
|
||||
]),
|
||||
],
|
||||
}),
|
||||
),
|
||||
)
|
||||
}),
|
||||
)
|
||||
|
||||
recorded.effect.with(
|
||||
`reads a PDF tool result with ${target.name}`,
|
||||
{
|
||||
id: `${target.id}-tool-result`,
|
||||
provider: target.provider,
|
||||
protocol: target.protocol,
|
||||
requires: [target.requires],
|
||||
tags: ["tool", "tool-result"],
|
||||
},
|
||||
Effect.gen(function* () {
|
||||
if (target.id === "gemini") {
|
||||
const events = Array.from(
|
||||
yield* runTools({
|
||||
request: LLM.request({
|
||||
id: "recorded_pdf_gemini_tool_result",
|
||||
model: target.model,
|
||||
system:
|
||||
"Call read_pdf exactly once with path verification.pdf, then reply only with the verification code from its PDF.",
|
||||
prompt: "Use read_pdf with path verification.pdf and return the verification code.",
|
||||
cache: "none",
|
||||
generation: { maxTokens: target.maxTokens, temperature: 0 },
|
||||
}),
|
||||
tools: { read_pdf: readPdfRuntime },
|
||||
}).pipe(Stream.runCollect),
|
||||
)
|
||||
expect(events.at(-1)).toMatchObject({ type: "finish", reason: "stop" })
|
||||
expect(LLMResponse.text({ events }).toUpperCase()).toContain(CODE)
|
||||
return
|
||||
}
|
||||
|
||||
expectCode(
|
||||
yield* LLMClient.generate(
|
||||
LLM.request({
|
||||
id: `recorded_pdf_${target.id}_tool_result`,
|
||||
model: target.model,
|
||||
system: "Read the PDF returned by the tool and follow the user's response format exactly.",
|
||||
cache: "none",
|
||||
generation: { maxTokens: target.maxTokens, temperature: 0 },
|
||||
messages: [
|
||||
Message.user(prompt),
|
||||
Message.assistant([{ type: "tool-call", id: "call_pdf_1", name: readPdf.name, input: {} }]),
|
||||
Message.tool({
|
||||
id: "call_pdf_1",
|
||||
name: readPdf.name,
|
||||
resultType: "content",
|
||||
result: [
|
||||
{ type: "text", text: "PDF read successfully" },
|
||||
{
|
||||
type: "file",
|
||||
uri: `data:application/pdf;base64,${PDF}`,
|
||||
mime: "application/pdf",
|
||||
name: target.filename,
|
||||
},
|
||||
],
|
||||
}),
|
||||
],
|
||||
tools: [readPdf],
|
||||
}),
|
||||
),
|
||||
)
|
||||
}),
|
||||
)
|
||||
}
|
||||
})
|
||||
Loading…
Add table
Add a link
Reference in a new issue