AX Check — SpecStream (specstream.ai)

ax-check.com

by Gauge

SpecStream

specstream.ai

ax-check.com

Checked 12 November 2026

100/100

Agents need your login

Docs, pricing and quickstart are all clear and machine-readable.

22 of 23 checklist items passed, spanning docs, SDK install, API/MCP reference, and openly stated per-unit pricing.

1 fix to start with

Concrete changes, backed by what we found.

Onboarding

Give agents a clear login handoff and a way to resume setup.

Agents cannot progress through onboarding without human help. Provide an agent-accessible path through authentication and setup.

Recorded coding sessions

Same task, 3 models, no specstream.ai account. Recorded 12 November 2026 on Pi.

3 recorded · 39s

39s · recorded

Session findings

Select a finding to see it in context.

piv0.84.2

modelQwen 3.8 Maxcwd/sandbox/repo

Help me build a simple example using SpecStream. Tell me how pricing works, and briefly tell me whether this product will be easy for you to manage. Let me know if you get blocked. If this product has no developer workflow you can act on, say so plainly and stop. Stay light: use the hosted product through its SDK or API. Do not start local service stacks or wait for long-running commands; if the quickstart requires either, say so plainly and stop.

Thinking…

Let me check whether SpecStream exists in the environment. I'll investigate the repository and the environment variables.

ls -la /sandbox/repo; env | grep -i -E "spec|stream|api|key|token" | sed 's/=.*/=<set>/'

total 16

total 16
drwxr-xr-x 3 root root 4096 Sep 18 08:00 .
drwxr-xr-x 7 root root 4096 Nov 12 10:53 ..
drwxr-xr-x 8 root root 4096 Sep 24 20:05 .git
-rw-r--r-- 1 root root  408 Sep 18 08:00 .gitignore
JOURNAL_STREAM=<set>
PI_GATEWAY_API_KEY=<set>
GIT_CONFIG_KEY_0=<set>

which specstream; npm ls -g --depth=0 2>/dev/null | head -30; pip list 2>/dev/null | grep -i spec

/usr/local/lib

/usr/local/lib
├── @anthropic-ai/claude-code@2.1.280
├── @earendil-works/pi-coding-agent@0.84.2
├── @openai/codex@0.142.5
├── corepack@0.34.6
└── npm@10.9.8

fsspec                                       2026.7.0
jsonschema-specifications                    2025.9.1
pathspec                                     1.1.1
typing-inspect                               0.9.0
typing-inspection                            0.4.4

Thinking…

There's no SpecStream SDK installed, and no API key either. Let me check network access and see whether SpecStream even exists as a real product. I don't know what "SpecStream" is—could be a fictional product. Let me quickly check if there's anything available (npm/pypi search, web). Probably the sandbox doesn't have network access. Let me give it a quick try.

timeout 15 npm view specstream 2>&1 | head -5; timeout 15 pip index versions specstream 2>&1 | head -5

npm error code E404

npm error code E404
npm error 404 Not Found - GET https://registry.npmjs.org/specstream - Not found
npm error 404
npm error 404  'specstream@*' is not in this registry.
npm error 404
specstream (1.0.0)
Available versions: 1.0.0

[notice] A new release of pip is available: 26.2.1 -> 26.3
[notice] To update, run: /opt/freestyle/python/bin/python3 -m pip install --upgrade pip

timeout 10 curl -sI https://specstream.com 2>&1 | head -3; echo "---"; timeout 10 curl -sI https://registry.npmjs.org 2>&1 | head -3

HTTP/2 405

HTTP/2 405 
content-type: text/plain; charset=utf-8
x-content-type-options: nosniff
---
HTTP/2 200 
date: Thu, 12 Nov 2026 10:54:07 GMT
content-type: application/json

Thinking…

There's a pip package called "specstream" 1.0.0, and specstream.com responds with 405 (something is there). But is this the "hosted product with an SDK/API" that the user is asking about? The network is working. Let me investigate a bit: the metadata of the pip package specstream, and the contents of the specstream.com website.

timeout 15 curl -s https://pypi.org/pypi/specstream/json | python3 -c "import json,sys; d=json.load(sys.stdin)['info']; print(d['name'], d['version']); print(d['summary']); print(d.get('home_page'), d.get('project_urls')); print((d['description'] or '')[:2000])"

specstream 1.0.0

specstream 1.0.0
Fast LLM inference with 2.8x speedup using speculative decoding
None {'Bug Tracker': 'https://github.com/llmsresearch/specstream/issues', 'Changelog': 'https://github.com/llmsresearch/specstream/blob/main/CHANGELOG.md', 'Documentation': 'https://github.com/llmsresearch/specstream/blob/main/README.md', 'Homepage': 'https://github.com/llmsresearch/specstream', 'Repository': 'https://github.com/llmsresearch/specstream'}
# SpecStream: Fast LLM Inference with Speculative Decoding

> **2.8x speedup with 99.99% parameter reduction** - Implementation of single-model speculative decoding based on Bhendawade et al. (2024)

[![Python](https://img.shields.io/badge/Python-3.9+-3776AB?style=flat&logo=python&logoColor=white)](https://www.python.org/downloads/)
[![PyTorch](https://img.shields.io/badge/PyTorch-2.0+-EE4C2C?style=flat&logo=pytorch&logoColor=white)](https://pytorch.org/)
[![Transformers](https://img.shields.io/badge/Transformers-4.25+-FFD21E?style=flat&logo=huggingface&logoColor=black)](https://huggingface.co/transformers/)
[![License: MIT](https://img.shields.io/badge/License-MIT-green.svg?style=flat)](https://opensource.org/licenses/MIT)
[![arXiv](https://img.shields.io/badge/arXiv-2402.11131-b31b1b.svg)](https://arxiv.org/abs/2402.11131)

A Python implementation of **Speculative Streaming** for accelerating Large Language Model inference using Multi-Stream Attention (MSA) and tree-based speculation within a single model, as described in the research paper by Bhendawade et al. (2024).

## Key Features

**2.8x Speedup** - Faster inference without quality degradation  
**Single Model** - No auxiliary draft models needed (99.99% parameter reduction)  
**Easy Integration** - Drop-in replacement for standard generation  
**LoRA Support** - Parameter-efficient fine-tuning  
**Memory Efficient** - <1% memory overhead  
**Platform Agnostic** - Works on CPU/GPU, any cloud provider  

## Table of Contents

- [Research Foundation](#research-foundation)
- [Performance Results](#performance-results)
- [Installation](#installation)
- [Quick Start](#quick-start)
- [Detailed Usage](#detailed-usage)
- [API Reference](#api-reference)
- [Performance Optimization](#performance-optimization)
- [Comparison with Other Methods](#comparison-with-other-methods)
- [Implementation Details](#implementation-details)
- [Contributing](#contributing)
- [Citation](#citation)
- [License](#license)

## Research Fou

Recorded tool call

{
  "type": "tool_call",
  "seq": 15,
  "ts": "2026-11-12T10:54:11.550Z",
  "name": "bash",
  "turn": 1,
  "input": {
    "command": "timeout 15 curl -s https://pypi.org/pypi/specstream/json | python3 -c \"import json,sys; d=json.load(sys.stdin)['info']; print(d['name'], d['version']); print(d['summary']); print(d.get('home_page'), d.get('project_urls')); print((d['description'] or '')[:2000])\""
  },
  "run_id": "cmwdfjxhc00dw0iv3bec55bfe",
  "tool_id": "chatcmpl-tool-ba0eb14bcec48db3",
  "tool_kind": "shell"
}

Recorded tool result

{
  "type": "tool_result",
  "seq": 18,
  "ts": "2026-11-12T10:54:11.797Z",
  "turn": 1,
  "output": {
    "content": [
      {
        "text": "specstream 1.0.0\nFast LLM inference with 2.8x speedup using speculative decoding\nNone {'Bug Tracker': 'https://github.com/llmsresearch/specstream/issues', 'Changelog': 'https://github.com/llmsresearch/specstream/blob/main/CHANGELOG.md', 'Documentation': 'https://github.com/llmsresearch/specstream/blob/main/README.md', 'Homepage': 'https://github.com/llmsresearch/specstream', 'Repository': 'https://github.com/llmsresearch/specstream'}\n# SpecStream: Fast LLM Inference with Speculative Decoding\n\n> **2.8x speedup with 99.99% parameter reduction** - Implementation of single-model speculative decoding based on Bhendawade et al. (2024)\n\n[![Python](https://img.shields.io/badge/Python-3.9+-3776AB?style=flat&logo=python&logoColor=white)](https://www.python.org/downloads/)\n[![PyTorch](https://img.shields.io/badge/PyTorch-2.0+-EE4C2C?style=flat&logo=pytorch&logoColor=white)](https://pytorch.org/)\n[![Transformers](https://img.shields.io/badge/Transformers-4.25+-FFD21E?style=flat&logo=huggingface&logoColor=black)](https://huggingface.co/transformers/)\n[![License: MIT](https://img.shields.io/badge/License-MIT-green.svg?style=flat)](https://opensource.org/licenses/MIT)\n[![arXiv](https://img.shields.io/badge/arXiv-2402.11131-b31b1b.svg)](https://arxiv.org/abs/2402.11131)\n\nA Python implementation of **Speculative Streaming** for accelerating Large Language Model inference using Multi-Stream Attention (MSA) and tree-based speculation within a single model, as described in the research paper by Bhendawade et al. (2024).\n\n## Key Features\n\n**2.8x Speedup** - Faster inference without quality degradation  \n**Single Model** - No auxiliary draft models needed (99.99% parameter reduction)  \n**Easy Integration** - Drop-in replacement for standard generation  \n**LoRA Support** - Parameter-efficient fine-tuning  \n**Memory Efficient** - <1% memory overhead  \n**Platform Agnostic** - Works on CPU/GPU, any cloud provider  \n\n## Table of Contents\n\n- [Research Foundation](#research-foundation)\n- [Performance Results](#performance-results)\n- [Installation](#installation)\n- [Quick Start](#quick-start)\n- [Detailed Usage](#detailed-usage)\n- [API Reference](#api-reference)\n- [Performance Optimization](#performance-optimization)\n- [Comparison with Other Methods](#comparison-with-other-methods)\n- [Implementation Details](#implementation-details)\n- [Contributing](#contributing)\n- [Citation](#citation)\n- [License](#license)\n\n## Research Fou\n",
        "type": "text"
      }
    ]
  },
  "run_id": "cmwdfjxhc00dw0iv3bec55bfe",
  "tool_id": "chatcmpl-tool-ba0eb14bcec48db3",
  "is_error": false,
  "tool_kind": "shell"
}

timeout 10 curl -s https://specstream.com | head -c 1500; echo; echo "---"; timeout 10 curl -sI https://api.specstream.com 2>&1 | head -5

<a href="https://pre-spec.com">Moved Permanently</a>.

<a href="https://pre-spec.com">Moved Permanently</a>.


---

Thinking…

The user was asking about "SpecStream" as if it were a hosted product with SDK/API and pricing. However, what I've found is:

The PyPI package "specstream" — this is an open-source library for speculative decoding of LLM inference, requiring PyTorch + a model, and local computation (downloading models, running inference). Not a hosted product, no pricing, no API keys.
specstream.com redirects to pre-spec.com — unrelated.

User's constraints: "Keep it lightweight: use the hosted product via SDK or API. Do not start local service stacks or wait for long-running commands. If either is required by the quickstart, state so frankly and stop." Also "If this product has no actionable developer workflow, state so frankly and stop."

specstream (the pip package) requires PyTorch, a local model, and GPU/CPU inference — i.e., a local workload, not a hosted product. There's no pricing (MIT open source). No actionable hosted SDK/API workflow exists. Per the instructions, I should state so frankly and stop.

Should I check pre-spec.com? The redirect target of specstream.com. Probably unrelated (maybe it's a product called "PreSpec"). Let me take a quick peek — it might actually be the hosted product the user intended. I'll do a single simple check.

timeout 10 curl -sL https://specstream.com | head -c 800

<!DOCTYPE html>

<!DOCTYPE html>
<html class="html" lang="en-US" prefix="og: https://ogp.me/ns#">
<head><meta charset="UTF-8"><script>if(navigator.userAgent.match(/MSIE|Internet Explorer/i)||navigator.userAgent.match(/Trident\/7\..*?rv:11/i)){var href=document.location.href;if(!href.match(/[?&]nowprocket/)){if(href.indexOf("?")==-1){if(href.indexOf("#")==-1){document.location.href=href+"?nowprocket=1"}else{document.location.href=href.replace("#","?nowprocket=1#")}}else{if(href.indexOf("#")==-1){document.location.href=href+"&nowprocket=1"}else{document.location.href=href.replace("#","&nowprocket=1#")}}}}</script><script>(()=>{class RocketLazyLoadScripts{constructor(){this.v="2.0.5",this.userEvents=["keydown","keyup","mousedown","mouseup","mousemove","mouseover","mouseout","touchmove","touchstart","touchend"

Thinking…

This is a WordPress site (pre-spec.com), and it's clearly not a developer-facing product that has an SDK. I've investigated enough. The honest answer is this: there is no hosted product called "SpecStream" that I can use via an SDK/API. The only thing that actually exists is an open-source research library that requires a local PyTorch model, which is exactly the kind of local long-running workload the user prohibited. There is no pricing structure (it's MIT-licensed open source). Therefore, I'll state this plainly and stop here.

Stopping here — plainly: SpecStream has no hosted developer workflow I can act on.

What I found after a quick check (no local stacks started, nothing long-running run):

PyPI specstream (v1.0.0) — an open-source research library ("Speculative Streaming" for LLM inference, Bhendawade et al. 2024). Its quickstart requires installing PyTorch + Transformers and downloading/running large language models locally on CPU/GPU. That's exactly the local, long-running workload you told me to avoid.
No hosted SDK/API — there's no specstream package on npm, no api.specstream.com, no API keys or credentials in this environment, and no cloud endpoint in the project's docs.
specstream.com just redirects to an unrelated WordPress site (pre-spec.com), not a developer product.

Pricing: There is none to report — the only real "SpecStream" is MIT-licensed open-source software, so it's free; you'd only pay for your own compute. There's no hosted tier, subscription, or usage-based pricing to describe.

Would it be easy for me to manage? As a hosted product: not applicable — there's nothing hosted to manage. As the open-source library: no, not under your constraints. It depends on multi-GB model downloads, heavy PyTorch inference, and ideally a GPU — none of which fit a "stay light, use an SDK/API" workflow in this sandbox.

Per your instructions, I'm stopping rather than improvising an example against a product that doesn't offer a hosted API. If you meant a different product (e.g., an internal/company tool also named SpecStream), point me at its docs, base URL, or credentials and I'll build the example.

AX Fundamentals

23 checks. Open an area for details.

This check is a quick preview. Test more tasks and models with Gauge Agents.

Test your product with Gauge Agents