1use anyhow::{Context as _, Result};
2use futures::{AsyncBufReadExt, AsyncReadExt, StreamExt, io::BufReader, stream::BoxStream};
3use http_client::{AsyncBody, HttpClient, HttpRequestExt, Method, Request as HttpRequest};
4use serde::{Deserialize, Serialize};
5use serde_json::Value;
6pub use settings::KeepAlive;
7use std::time::Duration;
8
9pub const OLLAMA_API_URL: &str = "http://localhost:11434";
10
11#[cfg_attr(feature = "schemars", derive(schemars::JsonSchema))]
12#[derive(Clone, Debug, Default, Serialize, Deserialize, PartialEq)]
13pub struct Model {
14 pub name: String,
15 pub display_name: Option<String>,
16 pub max_tokens: u64,
17 pub keep_alive: Option<KeepAlive>,
18 pub supports_tools: Option<bool>,
19 pub supports_vision: Option<bool>,
20 pub supports_thinking: Option<bool>,
21}
22
23fn get_max_tokens(name: &str) -> u64 {
24 /// Default context length for unknown models.
25 const DEFAULT_TOKENS: u64 = 4096;
26 /// Magic number. Lets many Ollama models work with ~16GB of ram.
27 /// Models that support context beyond 16k such as codestral (32k) or devstral (128k) will be clamped down to 16k
28 const MAXIMUM_TOKENS: u64 = 16384;
29
30 match name.split(':').next().unwrap() {
31 "granite-code" | "phi" | "tinyllama" => 2048,
32 "llama2" | "stablelm2" | "vicuna" | "yi" => 4096,
33 "aya" | "codegemma" | "gemma" | "gemma2" | "llama3" | "starcoder" => 8192,
34 "codellama" | "starcoder2" => 16384,
35 "codestral" | "dolphin-mixtral" | "llava" | "magistral" | "mistral" | "mixstral"
36 | "qwen2" | "qwen2.5-coder" => 32768,
37 "cogito" | "command-r" | "deepseek-coder-v2" | "deepseek-r1" | "deepseek-v3"
38 | "devstral" | "gemma3" | "gpt-oss" | "granite3.3" | "llama3.1" | "llama3.2"
39 | "llama3.3" | "mistral-nemo" | "phi3" | "phi3.5" | "phi4" | "qwen3" | "yi-coder" => 128000,
40 _ => DEFAULT_TOKENS,
41 }
42 .clamp(1, MAXIMUM_TOKENS)
43}
44
45impl Model {
46 pub fn new(
47 name: &str,
48 display_name: Option<&str>,
49 max_tokens: Option<u64>,
50 supports_tools: Option<bool>,
51 supports_vision: Option<bool>,
52 supports_thinking: Option<bool>,
53 ) -> Self {
54 Self {
55 name: name.to_owned(),
56 display_name: display_name
57 .map(ToString::to_string)
58 .or_else(|| name.strip_suffix(":latest").map(ToString::to_string)),
59 max_tokens: max_tokens.unwrap_or_else(|| get_max_tokens(name)),
60 keep_alive: Some(KeepAlive::indefinite()),
61 supports_tools,
62 supports_vision,
63 supports_thinking,
64 }
65 }
66
67 pub fn id(&self) -> &str {
68 &self.name
69 }
70
71 pub fn display_name(&self) -> &str {
72 self.display_name.as_ref().unwrap_or(&self.name)
73 }
74
75 pub fn max_token_count(&self) -> u64 {
76 self.max_tokens
77 }
78}
79
80#[derive(Serialize, Deserialize, Debug)]
81#[serde(tag = "role", rename_all = "lowercase")]
82pub enum ChatMessage {
83 Assistant {
84 content: String,
85 tool_calls: Option<Vec<OllamaToolCall>>,
86 #[serde(skip_serializing_if = "Option::is_none")]
87 images: Option<Vec<String>>,
88 thinking: Option<String>,
89 },
90 User {
91 content: String,
92 #[serde(skip_serializing_if = "Option::is_none")]
93 images: Option<Vec<String>>,
94 },
95 System {
96 content: String,
97 },
98 Tool {
99 tool_name: String,
100 content: String,
101 },
102}
103
104#[derive(Serialize, Deserialize, Debug)]
105#[serde(rename_all = "lowercase")]
106pub enum OllamaToolCall {
107 Function(OllamaFunctionCall),
108}
109
110#[derive(Serialize, Deserialize, Debug)]
111pub struct OllamaFunctionCall {
112 pub name: String,
113 pub arguments: Value,
114}
115
116#[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
117pub struct OllamaFunctionTool {
118 pub name: String,
119 pub description: Option<String>,
120 pub parameters: Option<Value>,
121}
122
123#[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
124#[serde(tag = "type", rename_all = "lowercase")]
125pub enum OllamaTool {
126 Function { function: OllamaFunctionTool },
127}
128
129#[derive(Serialize, Debug)]
130pub struct ChatRequest {
131 pub model: String,
132 pub messages: Vec<ChatMessage>,
133 pub stream: bool,
134 pub keep_alive: KeepAlive,
135 pub options: Option<ChatOptions>,
136 pub tools: Vec<OllamaTool>,
137 pub think: Option<bool>,
138}
139
140impl ChatRequest {
141 pub fn with_tools(mut self, tools: Vec<OllamaTool>) -> Self {
142 self.stream = false;
143 self.tools = tools;
144 self
145 }
146}
147
148// https://github.com/ollama/ollama/blob/main/docs/modelfile.md#valid-parameters-and-values
149#[derive(Serialize, Default, Debug)]
150pub struct ChatOptions {
151 pub num_ctx: Option<u64>,
152 pub num_predict: Option<isize>,
153 pub stop: Option<Vec<String>>,
154 pub temperature: Option<f32>,
155 pub top_p: Option<f32>,
156}
157
158#[derive(Deserialize, Debug)]
159pub struct ChatResponseDelta {
160 #[allow(unused)]
161 pub model: String,
162 #[allow(unused)]
163 pub created_at: String,
164 pub message: ChatMessage,
165 #[allow(unused)]
166 pub done_reason: Option<String>,
167 #[allow(unused)]
168 pub done: bool,
169 pub prompt_eval_count: Option<u64>,
170 pub eval_count: Option<u64>,
171}
172
173#[derive(Serialize, Deserialize)]
174pub struct LocalModelsResponse {
175 pub models: Vec<LocalModelListing>,
176}
177
178#[derive(Serialize, Deserialize)]
179pub struct LocalModelListing {
180 pub name: String,
181 pub modified_at: String,
182 pub size: u64,
183 pub digest: String,
184 pub details: ModelDetails,
185}
186
187#[derive(Serialize, Deserialize)]
188pub struct LocalModel {
189 pub modelfile: String,
190 pub parameters: String,
191 pub template: String,
192 pub details: ModelDetails,
193}
194
195#[derive(Serialize, Deserialize)]
196pub struct ModelDetails {
197 pub format: String,
198 pub family: String,
199 pub families: Option<Vec<String>>,
200 pub parameter_size: String,
201 pub quantization_level: String,
202}
203
204#[derive(Deserialize, Debug)]
205pub struct ModelShow {
206 #[serde(default)]
207 pub capabilities: Vec<String>,
208}
209
210impl ModelShow {
211 pub fn supports_tools(&self) -> bool {
212 // .contains expects &String, which would require an additional allocation
213 self.capabilities.iter().any(|v| v == "tools")
214 }
215
216 pub fn supports_vision(&self) -> bool {
217 self.capabilities.iter().any(|v| v == "vision")
218 }
219
220 pub fn supports_thinking(&self) -> bool {
221 self.capabilities.iter().any(|v| v == "thinking")
222 }
223}
224
225pub async fn complete(
226 client: &dyn HttpClient,
227 api_url: &str,
228 request: ChatRequest,
229) -> Result<ChatResponseDelta> {
230 let uri = format!("{api_url}/api/chat");
231 let request_builder = HttpRequest::builder()
232 .method(Method::POST)
233 .uri(uri)
234 .header("Content-Type", "application/json");
235
236 let serialized_request = serde_json::to_string(&request)?;
237 let request = request_builder.body(AsyncBody::from(serialized_request))?;
238
239 let mut response = client.send(request).await?;
240
241 let mut body = Vec::new();
242 response.body_mut().read_to_end(&mut body).await?;
243
244 if response.status().is_success() {
245 let response_message: ChatResponseDelta = serde_json::from_slice(&body)?;
246 Ok(response_message)
247 } else {
248 let body_str = std::str::from_utf8(&body)?;
249 anyhow::bail!(
250 "Failed to connect to API: {} {}",
251 response.status(),
252 body_str
253 );
254 }
255}
256
257pub async fn stream_chat_completion(
258 client: &dyn HttpClient,
259 api_url: &str,
260 api_key: Option<&str>,
261 request: ChatRequest,
262) -> Result<BoxStream<'static, Result<ChatResponseDelta>>> {
263 let uri = format!("{api_url}/api/chat");
264 let request = HttpRequest::builder()
265 .method(Method::POST)
266 .uri(uri)
267 .header("Content-Type", "application/json")
268 .when_some(api_key, |builder, api_key| {
269 builder.header("Authorization", format!("Bearer {api_key}"))
270 })
271 .body(AsyncBody::from(serde_json::to_string(&request)?))?;
272
273 let mut response = client.send(request).await?;
274 if response.status().is_success() {
275 let reader = BufReader::new(response.into_body());
276
277 Ok(reader
278 .lines()
279 .map(|line| match line {
280 Ok(line) => serde_json::from_str(&line).context("Unable to parse chat response"),
281 Err(e) => Err(e.into()),
282 })
283 .boxed())
284 } else {
285 let mut body = String::new();
286 response.body_mut().read_to_string(&mut body).await?;
287 anyhow::bail!(
288 "Failed to connect to Ollama API: {} {}",
289 response.status(),
290 body,
291 );
292 }
293}
294
295pub async fn get_models(
296 client: &dyn HttpClient,
297 api_url: &str,
298 api_key: Option<&str>,
299 _: Option<Duration>,
300) -> Result<Vec<LocalModelListing>> {
301 let uri = format!("{api_url}/api/tags");
302 let request = HttpRequest::builder()
303 .method(Method::GET)
304 .uri(uri)
305 .header("Accept", "application/json")
306 .when_some(api_key, |builder, api_key| {
307 builder.header("Authorization", format!("Bearer {api_key}"))
308 })
309 .body(AsyncBody::default())?;
310
311 let mut response = client.send(request).await?;
312
313 let mut body = String::new();
314 response.body_mut().read_to_string(&mut body).await?;
315
316 anyhow::ensure!(
317 response.status().is_success(),
318 "Failed to connect to Ollama API: {} {}",
319 response.status(),
320 body,
321 );
322 let response: LocalModelsResponse =
323 serde_json::from_str(&body).context("Unable to parse Ollama tag listing")?;
324 Ok(response.models)
325}
326
327/// Fetch details of a model, used to determine model capabilities
328pub async fn show_model(
329 client: &dyn HttpClient,
330 api_url: &str,
331 api_key: Option<&str>,
332 model: &str,
333) -> Result<ModelShow> {
334 let uri = format!("{api_url}/api/show");
335 let request = HttpRequest::builder()
336 .method(Method::POST)
337 .uri(uri)
338 .header("Content-Type", "application/json")
339 .when_some(api_key, |builder, api_key| {
340 builder.header("Authorization", format!("Bearer {api_key}"))
341 })
342 .body(AsyncBody::from(
343 serde_json::json!({ "model": model }).to_string(),
344 ))?;
345
346 let mut response = client.send(request).await?;
347 let mut body = String::new();
348 response.body_mut().read_to_string(&mut body).await?;
349
350 anyhow::ensure!(
351 response.status().is_success(),
352 "Failed to connect to Ollama API: {} {}",
353 response.status(),
354 body,
355 );
356 let details: ModelShow = serde_json::from_str(body.as_str())?;
357 Ok(details)
358}
359
360#[cfg(test)]
361mod tests {
362 use super::*;
363
364 #[test]
365 fn parse_completion() {
366 let response = serde_json::json!({
367 "model": "llama3.2",
368 "created_at": "2023-12-12T14:13:43.416799Z",
369 "message": {
370 "role": "assistant",
371 "content": "Hello! How are you today?"
372 },
373 "done": true,
374 "total_duration": 5191566416u64,
375 "load_duration": 2154458,
376 "prompt_eval_count": 26,
377 "prompt_eval_duration": 383809000,
378 "eval_count": 298,
379 "eval_duration": 4799921000u64
380 });
381 let _: ChatResponseDelta = serde_json::from_value(response).unwrap();
382 }
383
384 #[test]
385 fn parse_streaming_completion() {
386 let partial = serde_json::json!({
387 "model": "llama3.2",
388 "created_at": "2023-08-04T08:52:19.385406455-07:00",
389 "message": {
390 "role": "assistant",
391 "content": "The",
392 "images": null
393 },
394 "done": false
395 });
396
397 let _: ChatResponseDelta = serde_json::from_value(partial).unwrap();
398
399 let last = serde_json::json!({
400 "model": "llama3.2",
401 "created_at": "2023-08-04T19:22:45.499127Z",
402 "message": {
403 "role": "assistant",
404 "content": ""
405 },
406 "done": true,
407 "total_duration": 4883583458u64,
408 "load_duration": 1334875,
409 "prompt_eval_count": 26,
410 "prompt_eval_duration": 342546000,
411 "eval_count": 282,
412 "eval_duration": 4535599000u64
413 });
414
415 let _: ChatResponseDelta = serde_json::from_value(last).unwrap();
416 }
417
418 #[test]
419 fn parse_tool_call() {
420 let response = serde_json::json!({
421 "model": "llama3.2:3b",
422 "created_at": "2025-04-28T20:02:02.140489Z",
423 "message": {
424 "role": "assistant",
425 "content": "",
426 "tool_calls": [
427 {
428 "function": {
429 "name": "weather",
430 "arguments": {
431 "city": "london",
432 }
433 }
434 }
435 ]
436 },
437 "done_reason": "stop",
438 "done": true,
439 "total_duration": 2758629166u64,
440 "load_duration": 1770059875,
441 "prompt_eval_count": 147,
442 "prompt_eval_duration": 684637583,
443 "eval_count": 16,
444 "eval_duration": 302561917,
445 });
446
447 let result: ChatResponseDelta = serde_json::from_value(response).unwrap();
448 match result.message {
449 ChatMessage::Assistant {
450 content,
451 tool_calls,
452 images: _,
453 thinking,
454 } => {
455 assert!(content.is_empty());
456 assert!(tool_calls.is_some_and(|v| !v.is_empty()));
457 assert!(thinking.is_none());
458 }
459 _ => panic!("Deserialized wrong role"),
460 }
461 }
462
463 #[test]
464 fn parse_show_model() {
465 let response = serde_json::json!({
466 "license": "LLAMA 3.2 COMMUNITY LICENSE AGREEMENT...",
467 "details": {
468 "parent_model": "",
469 "format": "gguf",
470 "family": "llama",
471 "families": ["llama"],
472 "parameter_size": "3.2B",
473 "quantization_level": "Q4_K_M"
474 },
475 "model_info": {
476 "general.architecture": "llama",
477 "general.basename": "Llama-3.2",
478 "general.file_type": 15,
479 "general.finetune": "Instruct",
480 "general.languages": ["en", "de", "fr", "it", "pt", "hi", "es", "th"],
481 "general.parameter_count": 3212749888u64,
482 "general.quantization_version": 2,
483 "general.size_label": "3B",
484 "general.tags": ["facebook", "meta", "pytorch", "llama", "llama-3", "text-generation"],
485 "general.type": "model",
486 "llama.attention.head_count": 24,
487 "llama.attention.head_count_kv": 8,
488 "llama.attention.key_length": 128,
489 "llama.attention.layer_norm_rms_epsilon": 0.00001,
490 "llama.attention.value_length": 128,
491 "llama.block_count": 28,
492 "llama.context_length": 131072,
493 "llama.embedding_length": 3072,
494 "llama.feed_forward_length": 8192,
495 "llama.rope.dimension_count": 128,
496 "llama.rope.freq_base": 500000,
497 "llama.vocab_size": 128256,
498 "tokenizer.ggml.bos_token_id": 128000,
499 "tokenizer.ggml.eos_token_id": 128009,
500 "tokenizer.ggml.merges": null,
501 "tokenizer.ggml.model": "gpt2",
502 "tokenizer.ggml.pre": "llama-bpe",
503 "tokenizer.ggml.token_type": null,
504 "tokenizer.ggml.tokens": null
505 },
506 "tensors": [
507 { "name": "rope_freqs.weight", "type": "F32", "shape": [64] },
508 { "name": "token_embd.weight", "type": "Q4_K_S", "shape": [3072, 128256] }
509 ],
510 "capabilities": ["completion", "tools"],
511 "modified_at": "2025-04-29T21:24:41.445877632+03:00"
512 });
513
514 let result: ModelShow = serde_json::from_value(response).unwrap();
515 assert!(result.supports_tools());
516 assert!(result.capabilities.contains(&"tools".to_string()));
517 assert!(result.capabilities.contains(&"completion".to_string()));
518 }
519
520 #[test]
521 fn serialize_chat_request_with_images() {
522 let base64_image = "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNk+M9QDwADhgGAWjR9awAAAABJRU5ErkJggg==";
523
524 let request = ChatRequest {
525 model: "llava".to_string(),
526 messages: vec![ChatMessage::User {
527 content: "What do you see in this image?".to_string(),
528 images: Some(vec![base64_image.to_string()]),
529 }],
530 stream: false,
531 keep_alive: KeepAlive::default(),
532 options: None,
533 think: None,
534 tools: vec![],
535 };
536
537 let serialized = serde_json::to_string(&request).unwrap();
538 assert!(serialized.contains("images"));
539 assert!(serialized.contains(base64_image));
540 }
541
542 #[test]
543 fn serialize_chat_request_without_images() {
544 let request = ChatRequest {
545 model: "llama3.2".to_string(),
546 messages: vec![ChatMessage::User {
547 content: "Hello, world!".to_string(),
548 images: None,
549 }],
550 stream: false,
551 keep_alive: KeepAlive::default(),
552 options: None,
553 think: None,
554 tools: vec![],
555 };
556
557 let serialized = serde_json::to_string(&request).unwrap();
558 assert!(!serialized.contains("images"));
559 }
560
561 #[test]
562 fn test_json_format_with_images() {
563 let base64_image = "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNk+M9QDwADhgGAWjR9awAAAABJRU5ErkJggg==";
564
565 let request = ChatRequest {
566 model: "llava".to_string(),
567 messages: vec![ChatMessage::User {
568 content: "What do you see?".to_string(),
569 images: Some(vec![base64_image.to_string()]),
570 }],
571 stream: false,
572 keep_alive: KeepAlive::default(),
573 options: None,
574 think: None,
575 tools: vec![],
576 };
577
578 let serialized = serde_json::to_string(&request).unwrap();
579
580 let parsed: serde_json::Value = serde_json::from_str(&serialized).unwrap();
581 let message_images = parsed["messages"][0]["images"].as_array().unwrap();
582 assert_eq!(message_images.len(), 1);
583 assert_eq!(message_images[0].as_str().unwrap(), base64_image);
584 }
585}