एक providers.yaml से Raycast AI में catalog models डालें।
Updated 2026-07-30
Raycast की Custom Providers feature एक providers.yaml file के through कोई भी OpenAI-compatible endpoint accept करती है: base_url, एक key, और आपके declare किए models। फिर Claude, GPT, Gemini, और DeepSeek ids launcher के model picker में बैठते हैं, एक ही key से billed।
Quick answer: Custom Providers enable करें, एक file edit करें।
OpenAI-compatible endpoints के लिए Raycast का path Custom Providers feature है, advanced users के लिए aimed और default में disabled। इसे Raycast Settings में AI section के नीचे enable करें, config folder खोलने के लिए Reveal Providers Config इस्तेमाल करें, और shipped providers.template.yaml को providers.yaml में copy करें। File ~/.config/raycast/ai/providers.yaml पर रहती है। हर provider entry एक id, एक display name, एक base_url, और एक api_keys block लेती है; picker में चाहिए हर model explicitly declared है इसकी id, एक display name, और इसकी context window के साथ, plus एक abilities block जो describe करता है Raycast उससे क्या मांग सकता है। base_url shape built-in local-model examples वाला ही convention follow करती है, जो एक /v1 root पर point करते हैं, तो APIsRouter की value https://api.apisrouter.com/v1 है। File में credentials रहती हैं, तो इसे किसी भी secrets file की तरह treat करें।
providers:
- id: apisrouter
name: APIsRouter
base_url: https://api.apisrouter.com/v1
api_keys:
default: sk-APIsRouter-...
models:
- id: claude-sonnet-4-6
name: Claude Sonnet 4.6
context: 200000
abilities:
temperature:
supported: true
tools:
supported: true
- id: claude-haiku-4-5-20251001
name: Claude Haiku 4.5
context: 200000
abilities:
temperature:
supported: trueदो Raycast features जो एक जैसे सुनाई देते हैं, और नहीं हैं।
Raycast अपनी AI access लाने के दो तरीके document करता है, और एक ढूंढना भरोसे से दूसरा सामने ले आता है, तो difference साफ़ बताना worth है। Bring Your Own Keys, Raycast manual का BYOK page, आपकी personal Anthropic, Google, या OpenAI key (iOS पर OpenRouter) को Raycast AI से connect करता है। यह simpler feature है, बिना Pro subscription के काम करने के तौर पर documented, लेकिन यह एक custom endpoint नहीं है: requests API unification के लिए Raycast के servers से route होती हैं, और manual explicit है कि सिर्फ वे models accessible हैं जो पहले से Raycast AI में हैं। एक gateway key वहां plug नहीं होती, क्योंकि BYOK कभी URL नहीं मांगता। Custom Providers वह feature है जो यह page configure करता है: आपका अपना base_url, आपकी अपनी key, आपके declare किए models, requests वहां जाती हैं जहां आप point करते हैं। यह एक multi-vendor gateway के लिए path है, local servers के लिए, और किसी भी model के लिए जो Raycast की built-in list नहीं रखती। Trade explicitness है, Raycast आपके लिए endpoint की model list fetch नहीं करता (वह convenience एक standing feature request है), तो picker exactly वही दिखाता है जो आपका YAML declare करता है, ना ज़्यादा ना कम।
Models को honestly declare करना: ids, context, abilities।
चूंकि कोई autodiscovery नहीं है, YAML एक contract है, और इसमें हर field असली काम करती है। Model id exactly gateway की /v1/models listing से match करनी चाहिए; यह वही है जो request में travel करती है। Name सिर्फ वह label है जो Raycast दिखाता है। Context value Raycast को बताती है कि यह एक request में कितनी conversation history pack कर सकता है, तो इसे understate करना capability waste करता है और overstate करना ऐसी requests produce करता है जिन्हें model reject करता है; जिस id को आप declare कर रहे हैं उसकी documented window इस्तेमाल करें। Abilities block वह है जिसे लोग गलत करते हैं। यह declare करता है Raycast किस model पर क्या rely कर सकता है: temperature control, vision input, system messages, tool use, reasoning effort। एक ability declare करना जो model के पास नहीं है clean errors की बजाय Raycast features के अंदर confusing runtime failures produce करता है, और एक real ability को omit करना matching Raycast behavior को disable कर देता है। Minimal शुरू करें, उन models के लिए temperature plus tools जिन्हें आप AI extensions के साथ इस्तेमाल करेंगे, और जैसे ही आप इन्हें model की documentation के against confirm करें abilities add करें। एक community-maintained Raycast extension specifically इस file को एक UI से manage करने के लिए exist करता है, हर change से पहले automatic backups के साथ, जानने लायक अगर hand-by-hand YAML आपकी पसंद नहीं है। किसी भी तरह, Raycast disk से file पढ़ता है, तो edit करने के बाद, AI settings को एक moment दें या feature toggle करें यह confirm करने के लिए कि picker current file reflect करता है।
curl -s https://api.apisrouter.com/v1/models \
-H "Authorization: Bearer $APISROUTER_API_KEY" | head -50
# declare these ids verbatim in providers.yamlएक launcher के लिए models चुनना।
चूंकि हर declared model same key से bill होता है, comparison loop एक picker switch है: एक दिन के लिए दो ids पर same quick commands चलाएं, फिर console में per-model spend पढ़ें और उसे रखें जिसने अपनी जगह कमाई।
- Launcher AI burst work है: यह summarize करो, वह rewrite करो, selection explain करो। claude-haiku-4-5-20251001 और gemini-3.5-flash window animation ख़त्म होने से पहले return करते हैं, जो वह feel है जिसकी Raycast users उम्मीद रखते हैं।
- AI Chat sessions और लंबे drafting claude-sonnet-4-6 या gpt-5.5 कमाते हैं; इन्हें fast tier के साथ declare करें और picker में per task switch करें।
- AI extensions जो tools call करती हैं उन्हें एक ऐसा model चाहिए जो tool use पर भरोसेमंद हो, matching declared abilities block के साथ; claude-sonnet-4-6 वहां safe पहली choice है।
- deepseek-v4-flash उन users के लिए volume pick है जो हर text field में AI wire करते हैं; लगातार छोटे completions जुड़ते जाते हैं, और fast tier इस habit को balance पर invisible रखती है।
- बहुत सारे models speculatively की बजाय कुछ models deliberately declare करें: हर entry एक picker row है जिसे आप scroll करके निकालते हैं, और अगले id की ज़रूरत वाले दिन YAML को extend करना आसान है।
जितना उपयोग उतना भुगतान · आधिकारिक मूल्य से कम
Selected models are priced below official list prices. Exact input, output, cache, and per-request prices are shown for each model.
| मॉडल | आधिकारिक मूल्य | हमारा मूल्य |
|---|---|---|
| Claude Sonnet 4.6 | $3.00 / $15.00 per M | $2.40 / $12.00 per M |
| Claude Haiku 4.5 20251001 | $1.00 / $5.00 per M | $0.80 / $4.00 per M |
| GPT-5.5 | $5.00 / $30.00 per M | $4.00 / $24.00 per M |
| Gemini 3.5 Flash | $1.50 / $9.00 per M | $1.20 / $7.20 per M |
| DeepSeek V4 Flash | $0.14 / $0.28 per M | $0.10 / $0.30 per M |
Raycast के लिए specific failure modes।
Custom Providers चाहते हुए BYOK configure करना top-level गलती है, और यह आपकी गलती नहीं है: दोनों features एक search space share करते हैं। अगर आप जिस flow में हैं वह एक vendor key मांगता है पर कभी URL नहीं, तो आप BYOK में हैं, और gateway वहां fit नहीं बैठता। Settings, AI, और नीचे Custom Providers toggle पर वापस जाएं। File का ignore हो जाना आमतौर पर मतलब है feature toggle off है, file अब भी providers.template.yaml नाम की है, या YAML में एक syntax error है, जिस case में Raycast के पास load करने के लिए कुछ valid नहीं है और picker बस कोई custom models नहीं दिखाता। किसी और गहरी चीज़ का शक करने से पहले YAML validate करें। एक model जो कुछ Raycast features पर error देता है पर बाकी पर नहीं वह एक abilities mismatch है: tool-using AI extensions fail होती हैं जबकि plain chat काम करता है जब tools एक ऐसे model पर declare हुआ जिसके पास यह नहीं है, या कभी declare नहीं हुआ एक ऐसे model पर जिसके पास है। Size के लिए reject हुई requests एक overstated context value की तरफ इशारा करती हैं। और platform boundary को honestly नोट करें: Custom Providers Mac पर configure होता है, एक local config file में। अगर आपका कुछ Raycast इस्तेमाल कहीं और है, तो यह मान लेने से पहले कि parity है manual check करें feature वहां क्या support करता है।
कौन Raycast AI को एक gateway के through route करता है।
- Power users जो launcher में जीते हैं और quick AI commands fast catalog ids पर चाहते हैं बिना यह तय करने वाली एक subscription के कि वे कौन से models छू सकते हैं।
- People जो पहले से अपने editor और terminal tools एक gateway से route करते हैं और launcher को same key पर चाहते हैं, हर surface पर एक usage log।
- Users जो ऐसे models चाहते हैं जो Raycast की built-in list नहीं रखती, DeepSeek और GLM ids शामिल, YAML में एक बार declared और app में हर जगह उपलब्ध।
- AI-extension builders जिन्हें अपने extension के पीछे एक specific tool-capable model चाहिए, id से pinned ना कि एक hosted list के subject।
- Developers जिनके पास किसी given vendor की billing तक access नहीं है। बिना card requirement वाला top-up based access per-provider sign-up dependency हटा देता है।
Endpoint verify करें और पहली command को debug करें।
पहले models curl चलाएं और इसके output से ids को YAML में copy करें; memory से ids type करना यहां model-not-found errors की सबसे बड़ी वजह है, क्योंकि file ही Raycast का एकमात्र model source है। फिर toggle enable करें, confirm करें picker आपके declared names दिखाता है, और fast model पर एक quick AI command चलाएं। एक खाली picker toggle, filename, या YAML syntax है। एक authentication error api_keys block है। एक not-found error आपने अभी curl की listing के against एक id mismatch है। एक command जो chat में काम करती है पर एक AI extension में fail होती है वह उस model पर abilities declaration है। एक बार commands flow करने लगें, APIsRouter console per-request model, token counts, और spend दिखाता है। Launcher AI कुछ बड़ी requests की बजाय सैकड़ों छोटी requests हैं, और usage log वह जगह है जहां वह pattern एक number बन जाता है, per model, per day, उसी page पर जहां हर दूसरा tool है जो आप gateway से route करते हैं।
curl -s https://api.apisrouter.com/v1/chat/completions \
-H "Authorization: Bearer $APISROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"claude-haiku-4-5-20251001",
"messages":[{"role":"user","content":"ping"}]}'अक्सर पूछे जाने वाले प्रश्न
मैं Raycast AI में एक custom OpenAI-compatible endpoint कैसे add करूं?
Raycast की AI settings के नीचे Custom Providers enable करें, फिर ~/.config/raycast/ai/providers.yaml edit करें: base_url https://api.apisrouter.com/v1 और अपनी key वाली एक provider entry, plus id, name, और context के साथ explicit model declarations। Shipped providers.template.yaml आपके version के लिए schema document करती है।
क्या यह Raycast की Bring Your Own Keys जैसा ही है?
नहीं। BYOK एक personal Anthropic, Google, या OpenAI key connect करता है, Raycast के servers से route होता है, और सिर्फ वे models unlock करता है जो पहले से Raycast AI में हैं; यह कभी URL नहीं मांगता। Custom Providers वह file-based feature है जो एक base_url और आपकी अपनी model list लेता है, और एक gateway के लिए यही सही path है।
मेरे gateway models picker में क्यों नहीं दिखते?
Raycast custom endpoints से model lists fetch नहीं करता; picker exactly वही दिखाता है जो providers.yaml declare करता है। एक खाली picker मतलब Custom Providers toggle off है, file गलत नाम की है या इसमें invalid YAML है, या provider के तहत कोई models block declare नहीं हुआ।
abilities block क्या करता है?
यह declare करता है Raycast हर model से क्या मांग सकता है: temperature, vision, system messages, tools, reasoning effort। एक ability declare करना जो model के पास नहीं है उन features में confusing failures cause करता है जो इसे इस्तेमाल करती हैं, और एक real ability omit करना matching Raycast behavior disable कर देता है। Conservatively declare करें और confirm करते हुए expand करें।
क्या Custom Providers को Raycast Pro subscription चाहिए?
Raycast BYOK को बिना Pro के usable document करता है, और Custom Providers advanced users के लिए aimed एक settings toggle है। AI features के around plan gating समय के साथ shift हुई है, तो जिस हफ़्ते यह set करें उस current Raycast manual check करें आपका plan क्या include करता है।
क्या Raycast एक provider entry के through Claude, Gemini, और DeepSeek चला सकता है?
हां। हर declared model की id base_url को एक plain string के तौर पर forward होती है, तो एक provider entry claude-sonnet-4-6, gemini-3.5-flash, और deepseek-v4-flash side by side list कर सकती है, सब same key से billed और picker में switchable।