@bastani/atomic-ai 0.10.2-alpha.1 → 0.10.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md
CHANGED
|
@@ -4,6 +4,23 @@ This package is Atomic's fork of `@earendil-works/pi-ai`, published as `@bastani
|
|
|
4
4
|
|
|
5
5
|
## [Unreleased]
|
|
6
6
|
|
|
7
|
+
## [0.10.2] - 2026-11-20
|
|
8
|
+
|
|
9
|
+
### Added
|
|
10
|
+
|
|
11
|
+
- `appendSystemMessageUpdates()`, `resolveRequestTranscript()` and `followsToolResults()` build the request transcript for providers without mid-conversation system messages: the system messages that open the conversation fold into the leading prompt, and each later one is sent as a user message at its place in the transcript, so the prompt earlier requests sent never changes. `collapseSystemMessages()` and `resolveTranscript()` keep their behavior ([#6092](https://github.com/bastani-inc/atomic/pull/6092)).
|
|
12
|
+
- `UserMessage`, `AssistantMessage` and `ToolResultMessage` accept a request-time `cacheBreakpoint` hint. Anthropic, Amazon Bedrock and OpenAI-compatible Claude routes put a prompt-cache breakpoint at the end of the input sent up to that message, within the provider's limit of four breakpoints ([#6092](https://github.com/bastani-inc/atomic/pull/6092)).
|
|
13
|
+
- Assistant messages record `requestedThinkingBudget` (the budget the thinking level resolved to), `thinkingBudgetTokens` (the budget a budget-based Claude request sent) and `strictToolsDisabled` (the request sent preferred strict tools without strict schemas after an earlier rejection). `thinkingBudgetForLevel()` and `DEFAULT_THINKING_BUDGETS` are exported ([#6092](https://github.com/bastani-inc/atomic/pull/6092)).
|
|
14
|
+
|
|
15
|
+
### Fixed
|
|
16
|
+
|
|
17
|
+
- Fixed providers without mid-conversation system messages (Google Gemini and Vertex, Amazon Bedrock, Anthropic and OpenAI models without that capability, OpenAI-compatible endpoints and Mistral) rewriting the cached system prompt whenever the prompt changed after the first request, such as on a new day or after a project instructions edit, which made the provider bill the whole conversation again. Prompt changes are now sent after the history as their own text, framed as system prompt updates and never inside a tool result: after the user message before them, after the tool results in the same user message on Anthropic and Amazon Bedrock, as a user turn after the function responses on Gemini and Vertex, or as a user message after the tool messages on OpenAI-compatible endpoints, with the fixed assistant message "I have processed the tool results." between them where the endpoint rejects a user message directly after a tool message, so user and assistant turns still alternate. A conversation whose earlier requests already merged such changes into the system prompt is rewritten once more, on its first request after updating ([#6092](https://github.com/bastani-inc/atomic/pull/6092)).
|
|
18
|
+
- Fixed a tool schema rebuilt with its object keys in another order counting as a new tool definition, which rewrote the tool list on providers that send it at the top of the request ([#6092](https://github.com/bastani-inc/atomic/pull/6092)).
|
|
19
|
+
- Fixed budget-based Claude thinking near the context window, on the Anthropic API and Amazon Bedrock. When fitting the budget under the context safety margin would leave less than the smallest budget, the request is now sent without thinking, and the budget returns once it fits again. Before, it was sent with a budget below the provider's 1,024-token minimum, or with a 1,024-token budget at or above `max_tokens`, both of which the provider rejects ([#6092](https://github.com/bastani-inc/atomic/pull/6092)).
|
|
20
|
+
- Fixed Mistral rejecting a request with a user message directly after a tool message, as after input steered in while tools ran: the fixed assistant message "I have processed the tool results." now goes between them, as on OpenAI-compatible endpoints configured with `requiresAssistantAfterToolResult` ([#6092](https://github.com/bastani-inc/atomic/pull/6092)).
|
|
21
|
+
- Fixed Amazon Bedrock rejecting a request with two user messages in a row, as after input steered in while tools ran: adjacent user messages are now joined into one ([#6092](https://github.com/bastani-inc/atomic/pull/6092)).
|
|
22
|
+
- Fixed Anthropic and Amazon Bedrock retrying strict tool schemas on the first request of a resumed session after the session's strict tools had been rejected for exceeding Claude's limits. A recorded rejection, or a reply sent without strict schemas for that reason, now keeps them off ([#6092](https://github.com/bastani-inc/atomic/pull/6092)).
|
|
23
|
+
|
|
7
24
|
## [0.10.2-alpha.1] - 2026-11-20
|
|
8
25
|
|
|
9
26
|
### Added
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"schemaVersion":6,"generatedAt":"2026-11-20T23:
|
|
1
|
+
{"schemaVersion":6,"generatedAt":"2026-11-20T23:56:29.166Z","structureHash":"3734db31f484d1f3dfcaec5319edaab467019c4da43b0eabf9ee1ac5a1670c80","inputsHash":"2178bbe8b6af0cdd83df141d247c941cb3e1e58857fc89c2dec1cc56bece1663","files":{"amazon-bedrock.json":"1dab2ff833ddd537c2acaf4cb4861ba25f95834de024d6d432ddc5528c2f60ee","ant-ling.json":"628feab3ef6c7d80ce96f75999804e1d9f05fec3851e5302e30fff3b69db90f6","anthropic.json":"bcb41bae2d7b1f499903851cdadf3119cb58bf9c9fd434203a89f458746b2d06","azure.json":"dc8a960bcf64cdb78c5cda7173282477c2eecdf4ed99d11a72aa1c137a7914eb","baseten.json":"09ae9052506d3c0dbc312786167597ca5b02044a9cc8e7b16150757100f5eea6","cerebras.json":"2eb969b2956844f9a641d3de52c19f1791293c58f607282fb3b0ebfe85ff66f8","cloudflare-ai-gateway.json":"e20691466b294627f874be8d1bf42a537a102d79fce6aec4ad60d5285e29d50d","cloudflare-workers-ai.json":"2ef8c6ccb217a574efcd562da45f59966b07a52d7f0ce92de01523513e92753b","deepseek.json":"10a296fb3e898f7715c80890c8af0af4f5fd58f22dbcd6612fdc5d762157ccdc","fireworks.json":"7f373be387f753c38086b0feb00e2435cd39b7913ac3a9d1ba0039be160cdee8","github-copilot.json":"668bf9b01d7ad511883707fb0aa79e907972e2ad3792b1d60c8f11bc67cbb13a","google-vertex.json":"f73cb486e25b4867b16749473e29fc3fcd09110186c53b12ef54f89f674e15f2","google.json":"83b5bae1a9a34e5b0ea681daf61be2c4d66b365145b928dcd39edb08584cdde7","groq.json":"087df37206e9eed8b10f1e558e4c496289bef60074b7e8afba4f20fb102b4ec2","huggingface.json":"bcba6ea866be68cff9522da8de16291ecf53ae3a9b057b3f01b02933191d863c","kimi-coding.json":"7980156c11b7abf7667aab8b2e1d9c006106f65a7b64a9dc960b8a7891270079","meta.json":"6ccc391fc86586638b4a2f26420f0e4763fccce04d746665cbd55f477124bb3b","minimax-cn.json":"fa15d1f67a07046d38e191af9bd97ed922ba45a99a11af4ad281151cc6b762aa","minimax.json":"1e0459dda3858ea24151182106fc4179b95f43a0a8dffc1c3d2e7db4c2d9206e","mistral.json":"efac2474fc22cf48215bf79222e47df8186a1e880ddc0bed3e0fe7947bc6660e","moonshotai-cn.json":"1a15468b1edf9205956e4c5ea41fc4cacc14eabf00523b9143e23a38264f57f7","moonshotai.json":"f31cdf2e6c81a403c4cb0e6b7e759ceff2820f966b08dbec80cfafa3f80dfc63","nvidia.json":"fe1b0e33b5bc24756a3b3660674a783e3d4a70ace904a7beafafd45b1a668731","openai-codex.json":"109fcbd8a74d5728fbb4217143f422a95f7605631aefdfb9cd97c84b11c5441a","openai-decisions.json":"0c7d54fcbac5e7388872a85c0db7f336c5a57c82d403266e9ae04f87561daf4f","openai-images.json":"95e906730507ae8adcd49afa81fc21f7cee13c12842a71ba04ab62c59637bdb6","openai.json":"1d198edd3be3b530f1c39b09c38a3d5533e718a3c6179aef420ea9ede43a063e","opencode-go.json":"2988f82862123acb2726db889c28080935cad0fbb6ccbe18d39b24c88f1d8248","opencode.json":"6babfe2c2787565b0491c530d2df07c64de4ea3fbd98783d85aaac0290a50689","openrouter.json":"17af5bd382d3f9898182dfb80859635681f2105bcc60277e6a5fa521879e0597","qwen-token-plan-cn.json":"acdcd16ff61cb48856c60f3190e32574546ffb5a7f1fd7fe4c759ad83f217082","qwen-token-plan-individual-cn.json":"50e4cc19f732a3f57813af5da0b908ebf26ff907f09f0f32f64d6c102866fc66","qwen-token-plan-individual.json":"d48740c24107d42b952dd1c4fbbc09cb9d85c5f439ded3e92b7b462e97f8e38f","qwen-token-plan.json":"4018b952fae26a99eb296e2846c74ee4e919d4cd50ef0c564655b95dccffa203","together.json":"e3c6f7e9ac64a75e7e395e614f0fee85a45f8c34463dc8bcfe0e7141167451b1","typesafe.json":"938c89dfb6f636dce1483ec6acc739af4b881f61003d007807a74414eb2bf357","vercel-ai-gateway.json":"7fa8697bb43fe886bf2dea228c6f253929e002e50782bedba0af0e1c97619118","xai.json":"59f8a7bdf14e79ba1b0d0a37e56251bbf100420bdac7d6af5942508fdb990f11","xiaomi-token-plan-ams.json":"bfd33741da693bdb8e8ebd4e7b90c3e90820aa18432ad081a5a98db13c6bd296","xiaomi-token-plan-cn.json":"4b61d78feae3a926fda918c3fd3a9afaf2ca60c1bba0af8270a5398fb50a983e","xiaomi-token-plan-sgp.json":"b05a197ce24fd69fc9440f4ba929984ac7a0fefbe8083268734b422e721987ec","xiaomi.json":"495b5fddf3c3ec8904fecc15c4c19536f4961dce7b51512b6909d2df2e739824","zai-coding-cn.json":"cc9e0768e726ef55d2b7af853ac5f3585dc72575932cb38840a3877d18736b78","zai.json":"58e6642fdfb736f0b9a4a277c760c536a3cea66a030abbfb2615f7b2851d5c6f"}}
|
package/dist/utils/version.d.ts
CHANGED
|
@@ -1,2 +1,2 @@
|
|
|
1
|
-
export declare const PI_AI_VERSION = "0.10.2
|
|
1
|
+
export declare const PI_AI_VERSION = "0.10.2";
|
|
2
2
|
//# sourceMappingURL=version.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"version.d.ts","sourceRoot":"","sources":["../../src/utils/version.ts"],"names":[],"mappings":"AACA,eAAO,MAAM,aAAa,
|
|
1
|
+
{"version":3,"file":"version.d.ts","sourceRoot":"","sources":["../../src/utils/version.ts"],"names":[],"mappings":"AACA,eAAO,MAAM,aAAa,WAAW,CAAC"}
|
package/dist/utils/version.js
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"version.js","sourceRoot":"","sources":["../../src/utils/version.ts"],"names":[],"mappings":"AAAA,sGAAsG;AACtG,MAAM,CAAC,MAAM,aAAa,GAAG,
|
|
1
|
+
{"version":3,"file":"version.js","sourceRoot":"","sources":["../../src/utils/version.ts"],"names":[],"mappings":"AAAA,sGAAsG;AACtG,MAAM,CAAC,MAAM,aAAa,GAAG,QAAQ,CAAC","sourcesContent":["// The @bastani/atomic-ai package version. scripts/bump-version.ts keeps it in sync with package.json.\nexport const PI_AI_VERSION = \"0.10.2\";\n"]}
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@bastani/atomic-ai",
|
|
3
|
-
"version": "0.10.2
|
|
3
|
+
"version": "0.10.2",
|
|
4
4
|
"description": "Atomic's unified LLM API: providers, models, auth, streaming, tools, images and classification",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"main": "./dist/index.js",
|
|
@@ -90,8 +90,8 @@
|
|
|
90
90
|
"dependencies": {
|
|
91
91
|
"@anthropic-ai/sdk": "0.129.0",
|
|
92
92
|
"@aws-sdk/client-bedrock-runtime": "3.1127.0",
|
|
93
|
-
"@bastani/atomic-schema": "0.10.2
|
|
94
|
-
"@bastani/atomic-telemetry": "0.10.2
|
|
93
|
+
"@bastani/atomic-schema": "0.10.2",
|
|
94
|
+
"@bastani/atomic-telemetry": "0.10.2",
|
|
95
95
|
"@google/genai": "2.32.0",
|
|
96
96
|
"@smithy/node-http-handler": "4.13.0",
|
|
97
97
|
"http-proxy-agent": "9.1.0",
|