AMP, OpenRouter & Ollama: agents chalane ke mazeed tarike
Frontier coding tak (taqriban) free rasai ke teen mazeed raaste — AMP ke ad-supported credits, OpenRouter par set kiya gaya Claude Code, aur Ollama ke sath locally chalne wale models.
01AMP Code — frontier performance, ad-supported
AMP OpenCode ki tarah hai: kisi ek provider se affiliated nahi hai. Yeh ek agentic coder hai jo multiple providers se connect ho sakta hai aur alag alag analyses run kar sakta hai. Yeh terminal tool ke taur par zyada famous hai, lekin VS Code, Cursor ya Windsurf ke extension ke taur par bhi aata hai (jo waise bhi taqriban ek hi cheez hain).
Pehli nazar mein AMP Free aisa lagta hai jaise iski cost $10/day hai — lekin yeh iska ulat hai. Aapko ads dekhne ke badlay daily $10 ka model credit milta hai. Tech community ads par gussa hona pasand karti hai, lekin yeh waqai ek achhi deal hai: agar ads inoffensive hain (shayad useful bhi), to daily $10 ki frontier-level usage hasil karne se inkar karna mushkil hai.
Install aur launch karein
AMP ki install command VS Code terminal mein paste karein, phir amp type karein.
Log in / sign up
Pehli launch par log in karne ke liye browser khulta hai. Account nahi hai? Email/password ya Google auth se sign up karein — phir aap seedha terminal mein wapas aa jayein ge.
Ad notice karein
Top par ek ad dikhaya jata hai — is tarah aap $10/day ki free usage ke liye pay karte hain.
Model ke bajaye thinking mode pick karein
Ctrl+S se smart → deep → rush toggle hota hai. AMP aapke liye underlying model select karta hai; aap is par bharosa karte hain ki yeh task ke liye behtareen model choose karega.
Same code-review prompt ko deep mode mein chalane par, AMP ne ek critical
issue dhoond nikaala — phir se hard-coded credentials — aur ise fix karne ki offer ki. Context readout se pata chalta hai ki background mein ek high-end
OpenAI-class model hai. Is run par daily $10 mein se taqriban 42¢ kharch hue.
ampcode.com → settings par aap apna plan, remaining balance dekh sakte hain, credits buy kar sakte hain,
auto-top-up set kar sakte hain, aur GitHub connect kar sakte hain.
AMP aapko daily ad-supported allotment ke zariye free, frontier-level performance ka ek aasan rasta deta hai. Iska experience Claude Code jaisa hi hai — waise practice mein Claude Code ab bhi aage hai.
02Claude Code, OpenRouter par (Kimi K2)
Aap Claude Code ko khud kisi doosre provider ko call karne ke liye set kar sakte hain. Hum ise OpenRouter par Kimi K2.5 chalane ke liye point karein ge — jo shayad is waqt Moonshot AI ka sab se strong open-source model hai.
Claude Code ko Claude ke models ke sath chalane ke liye design aur optimise kiya gaya hai. Ise kahin aur redirect karna ek janky experience hai — yahan tak ki iski configuration bhi fiddly hai. Jab tak aap khas taur par experiment na karna chahein, follow karne ke bajaye dekhein; open-source kaam ke liye OpenCode ya AMP ko prefer karein.
Yeh trick environment variables ke ek set se hoti hai jo Anthropic tooling ko cloud mein Claude ke bajaye OpenRouter ko call karne ka batate hain. Aap default Haiku/Sonnet/Opus models ko apne chosen model par set karte hain (warna yeh wapas Claude par chala jata hai), ANTHROPIC_BASE_URL ko OpenRouter par point karte hain, apni OpenRouter
key ko auth token ke taur par use karte hain, aur real API key ko blank kar dete hain. Phir --model ke sath launch karein.
export ANTHROPIC_DEFAULT_HAIKU_MODEL="moonshotai/kimi-k2.5" export ANTHROPIC_DEFAULT_SONNET_MODEL="moonshotai/kimi-k2.5" export ANTHROPIC_DEFAULT_OPUS_MODEL="moonshotai/kimi-k2.5" export ANTHROPIC_BASE_URL="https://openrouter.ai/api" export ANTHROPIC_AUTH_TOKEN="<PUT YOUR OPENROUTER KEY HERE>" export ANTHROPIC_API_KEY="" claude --model moonshotai/kimi-k2.5
Launch karein, /model run karein, aur aap model ke taur par Moonshot AI Kimi K2 dekhein ge — yeh samajhta
hai ki iske paas doosre models bhi hain, lekin hum ne un sab ko Kimi K2 par map kar diya hai. Is se koi simple cheez poochein ("describe the
purpose of this project") aur yeh same Claude Code tooling use karke sahi jawab deta hai. Apne OpenRouter activity page par confirm karein ki recent calls waqai Kimi K2 par gayi hain.
Open-source models ke sath, kuch cheezein kaam karti hain aur kuch nahi; cheezein tezi se badalti hain (log fast Grok
variants bhi pasand karte hain). Agar free OpenRouter models aapko rate-limit kar dein, to model ko openrouter/free
par set karein taake aap kisi working free model par automatically route ho jayein.
03Ollama (GPT-OSS) ke sath locally chalana
Aakhir mein, hum Claude Code ko Ollama ke zariye aap ki apni machine par locally chalne wale model par point karte hain. Same idea — lekin ANTHROPIC_BASE_URL ab aap ke computer par http://localhost:11434 (Ollama ka default port) par point karta hai.
export ANTHROPIC_DEFAULT_HAIKU_MODEL="gpt-oss" export ANTHROPIC_DEFAULT_SONNET_MODEL="gpt-oss" export ANTHROPIC_DEFAULT_OPUS_MODEL="gpt-oss" export ANTHROPIC_BASE_URL="http://localhost:11434" export ANTHROPIC_AUTH_TOKEN="ollama" export ANTHROPIC_API_KEY="" ollama pull gpt-oss claude --model gpt-oss
Is se project summarise karne ko kahein aur — kafi achhi machine ke sath — yeh kaam karta hai: is ne local-only project-management MVP ko sahi identify kiya, hard-coded password dhoonda, aur ise chalane ka tarika jana. Aap GPU ko full speed par chalte hue dekh sakte hain, jo prove karta hai ki yeh locally chala. Lekin is ke liye ek heavy system chahiye.
Serious power ke liye, models ko locally chalana shayad hi kabhi behtareen rasta ho — OpenRouter bohot ziada strong models ko bohot cheap rate par serve kar sakta hai. Cloud GPU rent karna ek option hai, lekin aksar logon ke liye OpenRouter cost aur capability dono mein jeet jata hai.
Yeh environment variables sirf current terminal mein rehte hain. Un terminals ko close karein, ek new terminal kholein, aur aapka environment reset ho jata hai — aap wapas Claude ke sath Claude Code par aa jate hain, jahan hum kal ke liye hona chahte hain.
04Config recipes (reference)
Copy-paste recipes ka mukammal set. Key placeholder ko apni OpenRouter key se replace karein — kabhi bhi repo mein real key commit na karein.
$env:ANTHROPIC_DEFAULT_HAIKU_MODEL="moonshotai/kimi-k2.5" $env:ANTHROPIC_DEFAULT_SONNET_MODEL="moonshotai/kimi-k2.5" $env:ANTHROPIC_DEFAULT_OPUS_MODEL="moonshotai/kimi-k2.5" $env:ANTHROPIC_BASE_URL="https://openrouter.ai/api" $env:ANTHROPIC_AUTH_TOKEN="<PUT YOUR OPENROUTER KEY HERE>" $env:ANTHROPIC_API_KEY="" claude --model moonshotai/kimi-k2.5
$env:ANTHROPIC_DEFAULT_HAIKU_MODEL="gpt-oss" $env:ANTHROPIC_DEFAULT_SONNET_MODEL="gpt-oss" $env:ANTHROPIC_DEFAULT_OPUS_MODEL="gpt-oss" $env:ANTHROPIC_BASE_URL="http://localhost:11434" $env:ANTHROPIC_AUTH_TOKEN="ollama" $env:ANTHROPIC_API_KEY="" ollama pull gpt-oss claude --model gpt-oss
OpenCode: opencode.ai · AMP: ampcode.com · OpenRouter: openrouter.ai
Aaj ke liye platforms ke sath itna khilwaad kafi hai. Aap ne Claude Code, OpenCode aur AMP Code dekh liya hai, aur OpenRouter aur Ollama ke aganist Claude Code chalaya hai. Apne terminals reset karein aur aap wapas Claude ke sath Claude Code par aa jayein ge — jahan kal ke liye humein hona hai, kyun ke hum deep ja kar advanced users banein ge. Aap ab expert agentic engineer banne ke program mein 40% raasta tay kar chuke hain.
✓ Aham baatein
- AMP Free ads ke badlay frontier-level credit ka $10/day deta hai; yeh aap ke liye model select karta hai (Ctrl+S = smart/deep/rush).
- Claude Code ko environment variables ke zariye OpenRouter par redirect kiya ja sakta hai — yeh janky hai aur recommended nahi hai, lekin yeh kaam karta hai (Kimi K2.5).
- Ollama models ko locally chalata hai (GPT-OSS) agar aapki machine kafi heavy ho — lekin OpenRouter zyadatar cheaper aur stronger hai.
- Apni OpenRouter key ke liye hamesha placeholder use karein, aur real key kabhi commit na karein.
- Env vars clear karne aur standard Claude Code with Claude par wapas aane ke liye ek naya terminal kholein.