File size: 9,571 Bytes
eb1202c | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 | import { describe, expect, test } from "bun:test"
import { Effect } from "effect"
import { CacheHint, LLM, Message } from "../src"
import { Auth, LLMClient } from "../src/route"
import { AmazonBedrock } from "../src/providers"
import * as AnthropicMessages from "../src/protocols/anthropic-messages"
import * as Gemini from "../src/protocols/gemini"
import * as OpenAIChat from "../src/protocols/openai-chat"
import { applyCachePolicy } from "../src/cache-policy"
import { it } from "./lib/effect"
const anthropicModel = AnthropicMessages.route
.with({ endpoint: { baseURL: "https://api.anthropic.test/v1/" }, auth: Auth.header("x-api-key", "test") })
.model({ id: "claude-sonnet-4-5" })
const bedrockModel = AmazonBedrock.configure({
credentials: { region: "us-east-1", accessKeyId: "fixture", secretAccessKey: "fixture" },
}).model("anthropic.claude-3-5-sonnet-20241022-v2:0")
const openaiModel = OpenAIChat.route
.with({ endpoint: { baseURL: "https://api.openai.test/v1/" }, auth: Auth.bearer("test") })
.model({ id: "gpt-4o-mini" })
const geminiModel = Gemini.route
.with({
endpoint: { baseURL: "https://generativelanguage.test/v1beta/" },
auth: Auth.header("x-goog-api-key", "test"),
})
.model({ id: "gemini-2.5-flash" })
describe("applyCachePolicy", () => {
it.effect("undefined cache resolves to 'auto' (the recommended default)", () =>
Effect.gen(function* () {
const prepared = yield* LLMClient.prepare(
LLM.request({
model: anthropicModel,
system: "You are concise.",
prompt: "hi",
}),
)
// No explicit cache field → auto policy fires → last system part + latest
// user message both get cache_control markers.
expect(prepared.body).toMatchObject({
system: [{ type: "text", text: "You are concise.", cache_control: { type: "ephemeral" } }],
messages: [{ role: "user", content: [{ type: "text", text: "hi", cache_control: { type: "ephemeral" } }] }],
})
}),
)
it.effect("'auto' marks the last tool, last system part, and latest user message on Anthropic", () =>
Effect.gen(function* () {
const prepared = yield* LLMClient.prepare(
LLM.request({
model: anthropicModel,
system: "Sys A",
tools: [{ name: "t1", description: "t1", inputSchema: { type: "object", properties: {} } }],
messages: [
Message.user("first user"),
Message.assistant("assistant reply"),
Message.user("latest user message"),
],
cache: "auto",
}),
)
expect(prepared.body).toMatchObject({
tools: [{ name: "t1", cache_control: { type: "ephemeral" } }],
system: [{ type: "text", text: "Sys A", cache_control: { type: "ephemeral" } }],
messages: [
{ role: "user", content: [{ type: "text", text: "first user" }] },
{ role: "assistant", content: [{ type: "text", text: "assistant reply" }] },
{
role: "user",
content: [{ type: "text", text: "latest user message", cache_control: { type: "ephemeral" } }],
},
],
})
}),
)
it.effect("'auto' is a no-op on OpenAI (implicit caching protocol)", () =>
Effect.gen(function* () {
const prepared = yield* LLMClient.prepare(
LLM.request({
model: openaiModel,
system: "Sys",
prompt: "hi",
cache: "auto",
}),
)
const body = prepared.body as { messages: Array<{ content: unknown }> }
// OpenAI doesn't accept cache_control on messages — policy must skip.
const flat = JSON.stringify(body)
expect(flat).not.toContain("cache_control")
expect(flat).not.toContain("cachePoint")
}),
)
it.effect("'auto' is a no-op on Gemini (out-of-band caching protocol)", () =>
Effect.gen(function* () {
const prepared = yield* LLMClient.prepare(
LLM.request({
model: geminiModel,
system: "Sys",
prompt: "hi",
cache: "auto",
}),
)
const flat = JSON.stringify(prepared.body)
expect(flat).not.toContain("cache_control")
expect(flat).not.toContain("cachePoint")
}),
)
it.effect("'auto' on Bedrock emits cachePoint markers in the right places", () =>
Effect.gen(function* () {
const prepared = yield* LLMClient.prepare(
LLM.request({
model: bedrockModel,
system: "Sys",
tools: [{ name: "t1", description: "t1", inputSchema: { type: "object", properties: {} } }],
messages: [Message.user("first user"), Message.assistant("reply"), Message.user("latest user")],
cache: "auto",
}),
)
expect(prepared.body).toMatchObject({
toolConfig: {
tools: [{ toolSpec: { name: "t1" } }, { cachePoint: { type: "default" } }],
},
system: [{ text: "Sys" }, { cachePoint: { type: "default" } }],
messages: [
{ role: "user", content: [{ text: "first user" }] },
{ role: "assistant", content: [{ text: "reply" }] },
{ role: "user", content: [{ text: "latest user" }, { cachePoint: { type: "default" } }] },
],
})
}),
)
it.effect("'none' disables auto placement even when manual hints exist", () =>
Effect.gen(function* () {
const prepared = yield* LLMClient.prepare(
LLM.request({
model: anthropicModel,
system: "Sys",
tools: [{ name: "t1", description: "t1", inputSchema: { type: "object", properties: {} } }],
prompt: "hi",
cache: "none",
}),
)
expect(prepared.body).toMatchObject({
tools: [{ name: "t1", cache_control: undefined }],
system: [{ type: "text", text: "Sys", cache_control: undefined }],
})
}),
)
it.effect("granular object form: tools-only marks just tools", () =>
Effect.gen(function* () {
const prepared = yield* LLMClient.prepare(
LLM.request({
model: anthropicModel,
system: "Sys",
tools: [{ name: "t1", description: "t1", inputSchema: { type: "object", properties: {} } }],
prompt: "hi",
cache: { tools: true },
}),
)
expect(prepared.body).toMatchObject({
tools: [{ name: "t1", cache_control: { type: "ephemeral" } }],
system: [{ type: "text", text: "Sys", cache_control: undefined }],
})
}),
)
it.effect("auto policy preserves manual CacheHints on other parts", () =>
Effect.gen(function* () {
const prepared = yield* LLMClient.prepare(
LLM.request({
model: anthropicModel,
system: [
{ type: "text", text: "first system", cache: new CacheHint({ type: "ephemeral", ttlSeconds: 3600 }) },
{ type: "text", text: "last system" },
],
prompt: "hi",
cache: "auto",
}),
)
const body = prepared.body as { system: Array<{ text: string; cache_control?: unknown }> }
expect(body.system[0]?.cache_control).toEqual({ type: "ephemeral", ttl: "1h" })
expect(body.system[1]?.cache_control).toEqual({ type: "ephemeral" })
}),
)
it.effect("ttlSeconds in the policy flows through to wire markers", () =>
Effect.gen(function* () {
const prepared = yield* LLMClient.prepare(
LLM.request({
model: anthropicModel,
system: "Sys",
prompt: "hi",
cache: { system: true, ttlSeconds: 3600 },
}),
)
expect(prepared.body).toMatchObject({
system: [{ type: "text", text: "Sys", cache_control: { type: "ephemeral", ttl: "1h" } }],
})
}),
)
it.effect("messages: { tail: 2 } marks the last 2 message boundaries", () =>
Effect.gen(function* () {
const prepared = yield* LLMClient.prepare(
LLM.request({
model: anthropicModel,
messages: [Message.user("u1"), Message.assistant("a1"), Message.user("u2"), Message.assistant("a2")],
cache: { messages: { tail: 2 } },
}),
)
const body = prepared.body as { messages: Array<{ content: Array<{ cache_control?: unknown }> }> }
expect(body.messages[0]?.content[0]?.cache_control).toBeUndefined()
expect(body.messages[1]?.content[0]?.cache_control).toBeUndefined()
expect(body.messages[2]?.content[0]?.cache_control).toEqual({ type: "ephemeral" })
expect(body.messages[3]?.content[0]?.cache_control).toEqual({ type: "ephemeral" })
}),
)
it.effect("'latest-assistant' marks the last assistant message", () =>
Effect.gen(function* () {
const prepared = yield* LLMClient.prepare(
LLM.request({
model: anthropicModel,
messages: [Message.user("u1"), Message.assistant("a1"), Message.user("u2")],
cache: { messages: "latest-assistant" },
}),
)
const body = prepared.body as { messages: Array<{ content: Array<{ cache_control?: unknown }> }> }
expect(body.messages[0]?.content[0]?.cache_control).toBeUndefined()
expect(body.messages[1]?.content[0]?.cache_control).toEqual({ type: "ephemeral" })
expect(body.messages[2]?.content[0]?.cache_control).toBeUndefined()
}),
)
test("returns the same request reference when policy is a no-op (pure function)", () => {
const request = LLM.request({
model: anthropicModel,
prompt: "hi",
cache: "none",
})
expect(applyCachePolicy(request)).toBe(request)
})
})
|