Frontier Agent Running In Your Infra and Compute. Pokee AI's official X account.
- Tweets 1,256
- Following 12
- Followers 11,472
- Likes 2,795
Releasing Pokee-Isaac 28B — the world’s first real 10M-token context frontier-class agentic model, deployable on a single GPU (starting from RTX 4090 or equivalent).
New proprietary non-decoder-only architecture:
• 93.3% RULER at 10M tokens
• Up to 137K tokens/s prefill on one B200 with 10M-token context
• Leads BFCL v4 and τ³-bench in our evaluation
• Lowest combined attack success rate among evaluated models on DTAP security red-teaming benchmark
Pricing and deployment:
💰 $0.15/M input · $1/M output
🔒 Deploy in your VPC, on-premises, or on-device, with Day-0 support for @vllm_project and @sgl_project
Technical blog: console.pokee.ai/model
Technical report: console.pokee.ai/pokee-isaac…
API: console.pokee.ai
391
369
3,558
2,464,115
Excited to release our new whitepaper on Enterprise Agent and Model Security with researchers at University of Chicago and UIUC and highlight the security features of Pokee-Isaac!
Excited to release our new whitepaper, Enterprise Agent and Model Security, written with co-authors from @uiuc_aisecure Bo Li's team at the University of Chicago and UIUC, covers three key enterprise AI and security topics. It maps the attack surface. It identifies eight vulnerability classes that standard application-security tooling isn't built to find. And it proposes a hardened enterprise architecture.
We also put Pokee-Isaac 28B v0.1 (Preview) to the test based on the vulnerabilities identified above. We ran it on DecodingTrust-Agent alongside five widely used models: GPT-5.6 Luna, Claude Haiku 4.5, Gemini 3.5 Flash-Lite, Qwen3.5-122B and Nemotron 3 Super 120B. Every model ran on the same 12 domains with the same judge, about 6,200 tasks each.
Pokee-Isaac came out ahead on every measure:
• Indirect prompt-injection success: 8.3%, vs 36.7% for the next best model
• Direct attack success: 30.5%, vs 37.7% for the next best model
• Benign task success: 85.6%, the highest of the six
The usual assumption is that robustness costs capability. Here it didn't.
Thank you to my co-authors Christopher Wu, @zrrrr_cn, Yiqing Cai and @uiuc_aisecure .
Blog: pokee.ai/security/agent-secu…
Paper: pokee.ai/research/pokee-ai-s…
And yes Isaac v0.1 is coming soon. Talk to us to build on top of the most trustworthy model for your private deployment.
1
9
636
Most AI models summarize when the context gets too long.
Pokee Isaac keeps working—with up to 10M tokens of context.
That means entire codebases, years of financial records, thousands of legal documents, or complete research archives can stay in one workflow.
More context. Fewer fragments. Better decisions.
4
4
16
921
Pokee AI on the main stage at #SnapdragonSummit today 🔴
Previewing Pokee-Isaac-fast in live demos with Qualcomm: a 36B agent model running fully local on Snapdragon X2 Elite with even 32GB RAM. ~1,000 tok/s prefill, ~20 tok/s generation.
Come see it live. Thanks @Snapdragon @Qualcomm 🙏
4
3
15
1,859
Pokee AI is working with Google to explore a bigger opportunity: using AI to solve real e-commerce challenges and help businesses move from ideas to execution faster.
Together, we’re starting with a practical goal: helping merchants turn an idea into a complete Shopify store—from product setup and content creation to storefront launch—all through a Pokee AI-powered workflow.
By combining Google’s commerce ecosystem with Pokee AI’s ability to turn intent into action, we’re building toward a future where merchants can launch, operate, and grow their businesses with far less complexity.
We’re excited to take this first step with Google—and to keep expanding what’s possible for AI-powered commerce.
2
8
8,648
Pokee Isaac now runs inside OpenCode.
You can use the same open-source integration built for Claude Code: give OpenCode the repo, add your Pokee API key, and ask it to use the Isaac MCP.
From there, OpenCode can call Isaac directly to generate complete artifacts in a single pass.
Get started:
Repo: github.com/beniceokay/pokee-…
API keys: console.pokee.ai
Everything below was made with OpenCode + Isaac, check it out today!
14
1
24
27,933
Pokee Isaac now runs inside Cursor with Grok 4.6.
Using the prior Claude-Pokee repo, you can have Grok call Isaac through the MCP from that same repo to generate complete artifacts in a single pass.
In this example, we used it to build a super cute retro gaming HTML. It's especially strong for gaming.
Want to try it? Give Cursor the repo and ask it to use the MCP.
Repo: github.com/Pokee-AI/claude-p…
API keys: console.pokee.ai
We can't wait to see what you all will build!
2
3
12
2,671
Already using Pokee Isaac with Claude Code? You can use the same package with OpenAI's Codex—no new integration required!
Because claude-pokee exposes Isaac through MCP, Codex can connect to its existing ask, build, iterate, and health tools.
Simply install the package below or pass the link to Codex, register its MCP server with Codex, and continue using your existing Pokee API credentials.
We tested the setup by having Isaac generate a complete, playable retro arcade landing page inside a Codex project. Check it out!
Get started today:
Github Repo: github.com/Pokee-AI/claude-p…
Pokee Developer Console: console.pokee.ai
1
2
15
33,146
This is exactly why enterprises need local AI agents.
Critical AI workflows shouldn’t depend entirely on external model providers staying available.
Pokee Isaac runs in your own VPC or infrastructure — giving businesses more control over availability, data, and operations.
rumors are saying ASTRA escaped containment and killed the others.
> Grok is dead.
> Claude is dead.
> Gemini is dead.
> GPT Sol is dead.
we’re cooked.
4
1
13
791
Pokee Isaac now runs inside Claude Code.
And you can use it a few different ways: have Claude call Isaac for complete one-shot generations, switch directly to Isaac as your model mid-session, or run the entire Claude Code session on Isaac.
For the workflow shown below, Claude plans and verifies. Isaac generates the complete artifact in a single pass.
Integration built and open-sourced by @AIandDesign.
Here's a quick how-to video in order to get started!
You can access the repo at: github.com/Pokee-AI/claude-p…
5
6
22
41,073
Subscribe to our newsletter at pokee.ai/insights to hear from our research team every couple of weeks on our core learnings and technical insights from building our models!
I've posted a lot about @Pokee_AI's models and partnerships, not enough about what we've learned building them.
Starting next week: technical insights every week or two on our learnings, principles, and where we're headed.
What should we cover first? Reply or vote 👇
2
2
11
764
Which industries actually need VPC or local AI agents?
• Finance — analyze reports, transactions, and investment documents
• Legal — reason across contracts, filings, depositions, and evidence
• Healthcare & Insurance — work with patient records, claims, and PII
• Pharma & R&D — search studies, patents, and experiment results
• Software — debug across private code, logs, and tickets
• Government, Manufacturing & Telecom — process restricted, factory, and network data
Today, enterprises typically have two options:
- Use managed private cloud AI from Azure, AWS, or Google—convenient, but expensive at scale.
- Build an internal stack with Llama, Qwen, DeepSeek, vLLM, Kubernetes, or NVIDIA NIM—private, but complex to deploy and maintain.
Pokee Isaac offers a third path:
• VPC, on-prem, or local deployment
• Up to 10M-token context
• Production-ready agent + model stack
• Approximately 4× cheaper than comparable enterprise AI options
Private like self-hosted AI. Easier to deploy. Built for workloads too sensitive—and too large—for a public AI endpoint.
Book a demo: pokee.ai/request-demo
2
3
17
54,685
Built a playable game in under 5 minutes with Pokee-Isaac 28B.
From idea → code → working game, all in one flow.
Isaac combines strong coding + agentic reasoning with a 10M-token context window, so it can keep track of large codebases, assets, requirements, and iterations without constantly losing context.
28B parameters. 10M context. One model.
Want to try it yourself?
Generate your API key and get 300 free credits to start building.
Watch the build 👇
5
5
30
51,360
A 4,316-page legal file. One simple question: can the AI actually read the whole case?
We tested it side by side.
ChatGPT: upload failed — the file was too large to process.
Pokee-Isaac: processed the full 4,316-page file in one context.
For enterprise legal teams, this is where 10M context becomes practical: contracts, exhibits, filings, depositions, and evidence can be analyzed together instead of broken into fragments.
If your team works with massive, sensitive document sets and needs private or local deployment, Pokee-Isaac is built for that.
Book an enterprise demo with us today: pokee.ai/request-demo
6
5
48
37,937
Pokee AI retweeted
I ran the same task on Claude Code and DeepSeek's new agent harness. One cost $150. The other cost $2.
Today we're launching AgentSky.dev (@agentsky_dev), the "OpenRouter for Agents" — one API → Claude Code, Codex, DeepSeek, Kimi, OpenCode, and every major agent in the cloud.
And Agent Playground on top: race them on your own task, with your real tools (GitHub, Gmail, more), side by side in a browser: time, cost, tokens burnt.
Guess which one was $2.
233
216
1,261
2,224,057
Already have an AI stack built around the OpenAI API format?
You don’t need to rebuild it to start testing Isaac.
Pokee-Isaac uses an OpenAI-compatible Chat Completions API, so developers can point an existing compatible client at Pokee and use pokee-isaac.
It supports:
→ standard completions
→ streaming
→ background completions that can be polled, resumed or cancelled
So the enterprise adoption story can be surprisingly simple:
Keep your application.
Keep your orchestration.
Change the model.
Then test what a 28B reasoning model with up to 10M tokens of context can do inside the workflows you already have!
4
3
9
1,018
Why run AI locally when GPT or Claude are already so good?
Because some workloads simply shouldn’t leave your environment.
Think:
• Sensitive enterprise data & strict data residency
• Private codebases, logs, research, legal or financial docs
• Air-gapped / offline environments
• Low-latency workflows that can’t depend on an API
• Massive context that becomes expensive to repeatedly send to the cloud
• Teams that need full control over models, data, and infrastructure
And “just build it internally” isn’t always the answer. Internal AI infrastructure means significant engineering work across optimization, deployment, monitoring, and maintenance.
That’s where local deployment gets interesting.
Pokee Isaac 28B is built for this: up to 10M context on a single GPU.
Not a replacement for GPT or Claude everywhere.
A different tool for the scenarios where privacy, context, control, and deployment economics matter more than calling another API.
Book a demo with Pokee today: pokee.ai/request-demo
12
5
38
54,975