Agent Mesh
Wire models from different vendors into one graph. Let them debate a question and converge on a single answer, or run them independently and read the disagreement, which is often the useful part.
LIVE · v2.0 48 agents · 43 modules · 94 models
Built to collaborate with AI.
A room full of specialists, a graph you can wire models into, and 43 focused tools behind one workspace. Bring your own API keys and the models stay yours.
Claude Opus 5GPT-4.1Gemini 2.5 ProDeepSeek V3
Compare these three approaches and tell me which one survives contact with production.
CClaude Opus 5The second one. The first assumes ordered delivery, which the queue does not guarantee.
OGPT-4.1Agreed on the ordering point. I would add a retry budget before it ships.
Merged answer
Ship the queued approach with a bounded retry budget. Three of four models agree; the dissent is kept and attributed.
48Expert agents
43AI Center modules
27Model providers
94Models
43 modules, one set of keys
The work a chat window handles badly: speech, images, video, documents, markets and health, each with real controls instead of one prompt box. This is the row the app opens on.
AI ModelsEvery provider, one panel
Voice ChangerSpeak in another voice
Music SeparationSplit stems from a mix
AI SlidesA deck from an outline
Code ExplainerLine by line, in words
Math SolverWith the working shown
Stock Market AITickers, read carefully
Trend AnalyzerWhat is moving, and why
AI WritingDrafts in your register
Photo Upscaler4× without the mush
BG RemoverClean edges, one pass
AI VideoClips, dubs and subtitles
Logo & BrandMarks, palette, type
Translator60+ languages, in tone
AI Web StudioServices with no API
Podcast CreatorScript, voices, mixdown
Health DiagnosticsSignals, not diagnoses
Legal AdvisorClauses in plain words
Security AuditFindings, ranked
UI/UX GeneratorScreens with a system
SummarizerLong input, kept meaning
AI ResearchSources kept, not dropped
Prompt IDETune it against the model
SEO OptimizerIntent before keywords
E-Commerce AIListings that convert
Smart SchedulerPlans that survive a week
Sound GeneratorEffects from a sentence
AI PhotoGenerate, edit, restore
Every module runs on the keys you supplied, so a module you never open costs nothing. 10 of the 43 are open on the free tier.
Wire the models into one graph
Drop models onto a canvas, drag a cable between them and put them in a group. The group decides how they work: argue towards one answer, or answer alone so you can read where they disagree.
Add a model
Try “claude”, “gpt”…
CClaude−
Claude Opus 5 Flagship
Claude Fable 5 New
Claude Sonnet 5 1M ctx
Claude Haiku 4.5 Fast
OOpenAI+
GGemini−
Gemini 2.5 Pro Flagship
Gemini 2.5 Flash Free
DDeepSeek+
GGroq+
NNVIDIA NIM+
PPerplexity+
OGPT-4.1OpenAINo key · tap to add
GQwen3 32BGroqNo key · tap to add
CClaude Fable 5ClaudeNo key · tap to add
GGemini 2.5 ProGeminiNo key · tap to add
CClaude Sonnet 4.6ClaudeNo key · tap to add
CClaude Fable 5ClaudeNo key · tap to add
MLlama 3.1 405BGitHub Models
DDeepSeek V3DeepSeek
NNemotron 70BNVIDIA NIM
Claude Fable 5
Mode
sk-ant-… Save
Model
Claude Opus 5Complex agentic coding
Claude Fable 5The hardest reasoning tasks
Group
GPT-4.1Gemini 2.5 ProFable 5
Colour
Debate & convergeEach model reads the others and argues towards one answer.
Answer independentlyYou see every answer and choose.
Merged Chat
Any vendor, same canvas
Claude beside GPT beside Gemini beside a Llama on Groq. 94 models across 27 providers, each on its own key, wired together without leaving the page.
Groups are the unit of work
Give a group a shape, a colour and a mode. Every model you drop inside it inherits the brief, so a graph stays readable at fifteen nodes.
Debate, or answer alone
Debate & converge has each model read the others and argue towards one response. Answer independently keeps all of them and lets you pick.
One conversation at the end
Merged Chat collapses the whole graph into a single thread with every contribution still attributed, so you can see who said what.
Agent Office
48 specialists, a room each
An architect, a Flutter developer, a DevOps engineer, a QA lead, a designer, a data scientist, each with their own brief, their own thread and their own memory of your project.
- The thread stays. You explain your architecture once, to the agent who needs it, and it is still there next week.
- Departments, not a list. Management, development, design, data, security and operations, so you can brief a team rather than a person.
- They hand work to each other. The Strategic Coordinator delegates, collects the replies and reports back with one answer.
FLFlutter DeveloperMorpheus FlutterIdle
BEBackend DeveloperSixy YveWorking
ARSoftware ArchitectDaedalus PrimeWorking
UXUI/UX DesignerAphrodite PixelisIdle
QAQA EngineerCerberus VigilisIdle
DODevOps EngineerChronos PipelineWorking
AIAI EngineerAthena NeuralIdle
SCCyber SecuritySentinel ZeroIdle
Inside a module
Real controls, not a prompt box
A prompt box is the wrong instrument for speech, for upscaling or for a spreadsheet. Every module in AI Center exposes the parameters that actually change the result.
- Voice, language, pace, emotion as controls you can set and re-run, rather than adjectives you have to argue about in prose.
- Streaming output. The first audio arrives before the sentence is finished, so you hear a mistake early.
- The model is yours to choose. Each module lists the providers that can serve it and uses the key you gave it.
Categories
Creative Studio
Voice & Speech
Writing & Research
Code & Dev
Business
Social Media
Productivity
Health
Security & Legal
VoiceAria · warm
LanguageTurkish
ModelElevenLabs v3
Pace0.96×
Stability0.62
EmotionCalm
One workspace, not twelve tabs
Every part of Calik AI shares the same keys, the same context and the same account, so moving between them costs nothing.
Agent Office
48 specialists, a room each. Brief the one you need and keep the thread, instead of re-explaining your architecture to a general assistant every morning.
AI Center
43 focused modules for the work a chat window handles badly: images, speech, video, documents, research and analysis, each with real controls rather than a prompt box.
AI Chat
Every major model behind one conversation, on your own keys. Switch model mid-thread without losing what came before.
LLM Builder
Train on your own material, automatically or by demonstration. The result belongs to you, not to a platform.
PremiumVS Code Connect
The workspace reaches into your editor, so the agent reading your code is reading the file you actually have open.
Three steps, then it is yours
Bring your keys
Paste an API key for any of 27 providers, or several. Usage is billed by the provider at their price. Nothing is resold, and there is no per-token markup.
Pick the shape of the work
One model in AI Chat, a specialist in Agent Office, several wired together in Agent Mesh, or a purpose-built module in AI Center.
Keep the result
Conversations, trained models and generated work stay in your workspace and sync to your account, on every platform you sign in from.
Your keys, 27 providers
Usage is billed by the provider at their price. Nothing is resold and there is no per-token markup.
OpenAIAnthropicGoogle GeminiMeta LlamaMistralDeepSeekCoherePerplexityxAI GrokQwenMoonshot KimiZ.ai GLMTogether AIFireworks AIGroqNVIDIA NIMSambaNovaOpenRouterHyperbolicNebiusAI21 LabsGitHub ModelsCerebras
Free forever. Premium when you need it
Cancel any time. Yearly saves 20 per cent.
Free
Free
no card needed
no card needed
- 10 of the 43 AI Center modules
- 5 agents in Agent Office
- Agent Mesh with up to 2 agents
- All 27 providers, your own keys
- Five ways to get a key at no cost
Premium
$10
per month · $8 billed yearly
per month · $8 billed yearly
- All 43 modules and every agent
- Agent Mesh with no cap
- Advanced Prompt Booster
- Full LLM Builder
Ultra
Let’s talk
for teams
for teams
- Everything in Premium, every seat
- Shared workspaces and managed keys
- On-premise or private cloud
- SSO, audit logs, contractual SLA
Take Calik AI everywhere
One account, one set of keys, the same workspace. The web app runs everywhere today; the desktop builds add offline capability and the deeper editor integration.
Windows
Windows 10 and 11 · x64Native build, code-signed Open the workspace
macOS
macOS 10.15+ · UniversalApple silicon and Intel Open the workspace
Linux
x86_64 · tar.xz and AppImageGTK 3, no snap required Open the workspace
Android & iOS
Native mobile apps in buildSame account, same keys Use the web app
Built with Flutter, so all five targets come from one codebase and ship together. Until the mobile builds land, the web app is installable to a phone home screen and works the same way. Every desktop release is SHA-256 verified and code-signed.
Questions worth answering
Do I need my own API keys?
Yes, and that is the point. Every tier runs on your keys, so model usage is billed by the provider at their price rather than resold with a margin on top. Calik AI charges for the workspace, not for tokens.
What does Free actually include?
10 of the 43 AI Center modules, five agents in Agent Office, Agent Mesh with up to two agents, LLM Builder in automatic mode, and all 27 providers with your own keys. No card, and no expiry.
Can I cancel whenever I want?
Yes. Premium is month to month unless you choose yearly, and you can cancel at any time from Settings. Access runs to the end of the period you have already paid for.
Which platforms does it run on?
Windows, macOS, Linux and the web today, from one Flutter codebase. Native Android and iOS builds are in progress; until they ship, the web app installs to a phone home screen and behaves the same. Every desktop release is SHA-256 verified and code-signed.
Is my work used to train models?
No. Your conversations and files go to the provider whose key you supplied, under that provider's terms, and nowhere else. Calik AI does not train on your workspace.
What is the difference between Agent Mesh and Agent Office?
Agent Office is one specialist at a time, with a thread that remembers your project. Agent Mesh is several models at once on a canvas, wired into groups that either argue towards one answer or answer independently. Office is depth; Mesh is a second opinion, and a third.
Bring one key, open every model
Calik AI does not resell tokens. You connect a provider directly with a key that belongs to you, and the whole workspace runs on it. Twenty-seven providers are supported, and five of them hand out working keys without asking for a card.
Anthropic
Claude Opus 5, Fable 5, Sonnet 5 and Haiku 4.5. Prepaid credit rather than a free tier, and the strongest reasoning on this row.
sk-ant-
Google AI Studio
Gemini 2.5 Pro, 2.5 Flash and 2.0 Flash. A Google account is the whole requirement, and the free tier never asks for a payment method.
AIza
Groq
Llama 3.3 70B, Mixtral and Gemma2 on custom silicon. The fastest answers here, which is what makes it the one to put in Agent Mesh.
gsk_
Ollama
Llama 3, Mistral, Phi-3 and Qwen on your own machine. No key, no account, no request limit, and nothing you type crosses the network.
localhost:11434
GitHub Models and OpenRouter are free as well, and twenty-one more providers are supported: OpenAI, Grok, DeepSeek, Mistral, Qwen, Cohere, Perplexity, Azure OpenAI and the rest, plus OpenRouter as one key in front of a hundred models. The full list, with the steps for each and the key format the app expects, is on one page.
Get an API keyRead it before you commit to anything
Written to be worth reading whether or not you end up with an account here.
Setup
Get an API key
All 27 providers in order of how fast you can start, a full Anthropic Claude walkthrough, the five routes that cost nothing, and a table matching every common API error to its fix.
ModelsCompare models
Claude, GPT, Gemini, DeepSeek, Llama and Mistral by what each one is actually reached for. No benchmark tables, and no winner declared on your behalf.
Try itThe demo
Agent Mesh running, with a note at each stage saying what just happened. No key, no signup, nothing to install. Everything on it is recorded and says so.
EvidencePublished runs
Real Agent Mesh runs, published in full: one question, every model's answer beside it, and the points where they contradicted each other before converging.
$0
Margin on your model usage
5 of 27
Providers that need no card
4
Platforms today, mobile next
Start with the free tier.
No card, no expiry, and your own keys from the first minute. Upgrade when the caps start to matter, and cancel any time.
No card Your own keys No expiry on Free Cancel any time