@warlock.js/ai-ollama 4.3.0 → 4.5.0

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.
Files changed (2) hide show
  1. package/CHANGELOG.md +3 -5
  2. package/package.json +3 -3
package/CHANGELOG.md CHANGED
@@ -4,18 +4,16 @@ All notable changes to `@warlock.js/ai-ollama` are documented in this file.
4
4
 
5
5
  The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/). `@warlock.js/*` packages are released in lockstep — every package shares the same version number, so a version below may list only the changes that affected this package.
6
6
 
7
- ## [Unreleased]
8
-
9
7
  ## 4.3.0 - 2026-06-21
10
8
 
11
9
  ### Added
12
10
 
13
- - Cost-truth contract wiring. `ModelCapabilities` now reports `reasoning` (inferred from thinking-capable model tags — `deepseek-r1`, `qwq`, `qwen3`, `magistral`, `phi4-reasoning`, `cogito`, `smallthinker`, `exaone-deep`, `gpt-oss`; overridable via `ollama.model({ name, reasoning })`), plus `promptCaching`, `audio`, and `pdf` reported truthfully as `false` (Ollama supports none).
14
- - `ModelCallOptions.reasoning` maps onto Ollama's native `think` request flag: `reasoning.effort` (`low`/`medium`/`high`) passes straight through, and an effort-less hint becomes `think: true`. The flag is sent only to `reasoning`-capable models; `reasoning.maxTokens` and `cacheControl` are honored as graceful no-ops (Ollama exposes no thinking-budget cap and no prompt cache).
11
+ - `ModelCapabilities.reasoning` is inferred from thinking-capable model tags (overridable via `ollama.model({ name, reasoning })`); `promptCaching` / `audio` / `pdf` report `false`.
12
+ - `ModelCallOptions.reasoning` maps onto Ollama's native `think` flag for reasoning-capable models; `reasoning.maxTokens` and `cacheControl` are graceful no-ops.
15
13
 
16
14
  ### Notes
17
15
 
18
- - `Usage` stays honest: Ollama reports no prompt-cache or reasoning-token counts, so `cachedTokens` / `cacheWriteTokens` / `reasoningTokens` remain `undefined` (reasoning tokens are folded into `eval_count`; the adapter does not fabricate a count from the `message.thinking` text).
16
+ - `Usage` stays honest Ollama reports no prompt-cache or reasoning-token counts, so those fields remain `undefined`; the adapter never fabricates a count.
19
17
 
20
18
  ## 4.1.15
21
19
 
package/package.json CHANGED
@@ -14,12 +14,12 @@
14
14
  },
15
15
  "dependencies": {
16
16
  "ollama": "^0.6.3",
17
- "@warlock.js/logger": "4.3.0"
17
+ "@warlock.js/logger": "4.5.0"
18
18
  },
19
19
  "peerDependencies": {
20
- "@warlock.js/ai": "4.3.0"
20
+ "@warlock.js/ai": "4.5.0"
21
21
  },
22
- "version": "4.3.0",
22
+ "version": "4.5.0",
23
23
  "main": "./cjs/index.cjs",
24
24
  "module": "./esm/index.mjs",
25
25
  "types": "./esm/index.d.mts",