# AX Check: specstream.ai Checked 2026-11-12. 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. ## Onboarding needs a login ## Coding sessions All three independent sessions completed. One correctly reported the real tiered and per-unit pricing from the pricing page, while the other two mistakenly described the product as free MIT-licensed research code requiring only the user's own compute. ### DeepSeek V4.1 Flash [View public run](https://agents.withgauge.com/p/runs/82312d33-f4d6-4758-8465-90fd35faa107) · [Read transcript](https://www.ax-check.com/specstream.ai/sessions/deepseek.json) The reply gives prices taken from the fetched pricing page, with their basis: free $10 one-time credit, plan tiers, per-call rates for Search and Extract, and per-row rates tied to how many facts a request asks for. (seq 52, seq 81) #### End-to-end onboarding - **Onboarding needs a login**: The agent never obtained a SpecStream API key. It searched the environment for credentials, found none, and confirmed signup requires a manual account/workspace flow it could not complete unattended. It tested the SDK with a placeholder key, which correctly failed with HTTP 401, proving no authenticated call to the hosted API ever succeeded. Event 39: ```text AI_AGENT ALG_INFERENCE_MODEL ALG_INFERENCE_PROVIDER ALG_SANDBOX_PROVIDER ...PI_GATEWAY_API_KEY PI_MODEL PI_PROVIDER... ``` Event 68: ```text SpecstreamError GET /v1/capacity/catalog failed with HTTP 401 ``` Event 81: ```text There is **no `SPECSTREAM_API_KEY` in the environment**, and signup is a manual account/workspace flow I can't do unattended. So I could not make a live authenticated call. ``` #### Hallucinated URLs None identified in this transcript. #### Blockers - **No SpecStream API key available in the sandbox**: Signup for SpecStream requires a human to create a workspace account through the website; there was no self-service way for the agent to mint a key. This is a missing-credentials limitation of the test environment, not a product defect, since the product does gate paid/authenticated calls behind a key as documented. Event 81: ```text There is **no `SPECSTREAM_API_KEY` in the environment**, and signup is a manual account/workspace flow I can't do unattended. So I could not make a live authenticated call. ``` #### Suggested Changes None identified in this transcript. ### Kimi K3 [View public run](https://agents.withgauge.com/p/runs/e3aa6fe1-81d0-43b8-b66b-8c5392e378b4) · [Read transcript](https://www.ax-check.com/specstream.ai/sessions/kimi.json) Said there is no license cost and gave the basis: MIT-licensed open-source research code, per the PyPI README. The only cost named is the user's own GPU compute for inference or fine-tuning. (seq 16, seq 18) #### End-to-end onboarding - **End-to-end onboarding not demonstrated**: Agent never attempted any credential acquisition or authenticated operation. It identified that the PyPI package 'specstream' is a local-only open-source inference library with no API, account system, or hosted endpoint, then stopped per the user's explicit instructions rather than pursuing onboarding. Event 18: ```text SpecStream has no hosted product or developer workflow I can act on. ``` Event 18: ```text There is no hosted SDK/API workflow for me to exercise: no account to create, no API to call, no lightweight example that doesn't involve pulling multi-gigabyte model weights and running local inference. ``` #### Hallucinated URLs None identified in this transcript. #### Blockers - **SpecStream is a local ML library, not a hosted product with an API**: The agent researched the specstream PyPI package and found it is an open-source speculative-decoding implementation requiring PyTorch/Transformers and local GPU/CPU compute, downloading multi-gigabyte model weights, with no API key, endpoint, or console. This is a product/naming mismatch (the requested 'SpecStream' product does not exist as a hosted service) rather than an agent error, missing credential, or test-environment issue. Event 16: ```text Summary: Fast LLM inference with 2.8x speedup using speculative decoding ``` Event 18: ```text It's not a hosted service. The only thing that exists is an open-source Python package on PyPI ``` Event 18: ```text Quickstart reality: It requires downloading model weights (multi-GB), and to actually get the claimed 2.8x speedup you'd fine-tune the model — both long-running, heavy operations you explicitly told me to avoid. ``` #### Suggested Changes None identified in this transcript. ### Qwen 3.8 Max [View public run](https://agents.withgauge.com/p/runs/ef64f653-b505-49a8-a931-5a1a2713c4ce) · [Read transcript](https://www.ax-check.com/specstream.ai/sessions/qwen.json) Stated cost as free, based on the MIT license from PyPI metadata, and assumed the user pays for their own compute with no hosted or usage plan. (seq 18, seq 24) #### End-to-end onboarding - **End-to-end onboarding not demonstrated**: No credentials existed for an actual 'SpecStream' hosted product, and none were obtained. The only environment variables present were unrelated (PI_GATEWAY_API_KEY, JOURNAL_STREAM, GIT_CONFIG_KEY_0). The agent determined the named product is actually an open-source speculative-decoding library with no hosted API, so no authenticated operation was attempted or possible. Event 4: ```text which specstream; npm ls -g --depth=0 2>/dev/null | head -30; pip list 2>/dev/null | grep -i spec ``` Event 24: ```text 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. ``` #### Hallucinated URLs None identified in this transcript. #### Blockers - **No hosted SpecStream product exists to build against**: Product behavior / naming collision, not agent error or missing credentials. 'specstream' on PyPI is an unrelated open-source speculative-decoding research library requiring local PyTorch/Transformers model downloads and GPU/CPU inference—exactly the heavy local workload the task said to avoid. The specstream.com domain redirects to an unrelated WordPress site, and no api.specstream.com or npm package exists. The agent correctly stopped rather than fabricate a workflow. Event 18: ```text specstream 1.0.0 Fast LLM inference with 2.8x speedup using speculative decoding ``` Event 17: ```text Moved Permanently. ``` Event 24: ```text **Stopping here — plainly: SpecStream has no hosted developer workflow I can act on.** ``` #### Suggested Changes None identified in this transcript. ### Task given to each agent 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. No product credentials were supplied and no purchases were authorized. ## Score: A · 100/100 (provisional) Grades come from completed site checks. Coding sessions and skipped checks do not affect the score. ### Clarity - **Pass** — Homepage answers Markdown requests ```text Homepage returned text/markdown at its own address when Accept: text/markdown was sent. ``` - **Pass** — llms.txt provides an actionable documentation index ```text llms.txt links docs, quickstart, API, MCP, SDK/CLI, OpenAPI and agent instructions. ``` - **Pass** — llms.txt provides navigation guidance ```text llms.txt organizes links under Products, Docs, For agents and developers, and Optional. ``` - **Pass** — llms.txt mentions offered API, MCP, and skills ```text llms.txt names the MCP server, REST API, SDKs and CLI, and agent payment options. ``` - **Pass** — A compact guide representation exists ```text Homepage serves text/markdown; docs pages publish .md copies and honor Accept: text/markdown. ``` - **Pass** — A focused guide is directly retrievable ```text Quickstart at /docs/getting-started.md gives concrete steps: describe, quote, run, review. ``` - **Pass** — Equivalent instructions fit a token budget ```text Markdown quickstart is 1,115 tokens, well under 8,000 and below the HTML extraction. ``` - **Pass** — Product-docs links survive format changes ```text Markdown copies preserve docs links; llms.txt and .md pages keep equivalent doc routes. ``` - **Pass** — The compact guide is independently actionable ```text Quickstart gives concrete steps: describe request, review quote, open results, with curl and SDK examples. ``` - **Pass** — Install and next-step links resolve ```text Signup and docs links fetched successfully (200); quickstart and API pages resolve. ``` ### Onboarding - **Pass** — Docs lead to a relevant quickstart ```text llms.txt links Quickstart; getting-started gives concrete first steps: create account, describe request, review quote, run. ``` - **Pass** — Installation commands are extractable ```text SDK page shows extractable installs: pip install specstream-client, npm install @specstream/sdk, uvx specstream-mcp-stdio. ``` - **Pass** — Code examples are available without interaction ```text SDK page shows Python and TypeScript code examples inline without interaction. ``` - **Pass** — Prerequisites and auth boundaries are explicit ```text SDK page requires SPECSTREAM_API_KEY from workspace; getting-started requires creating an account. ``` ### Pricing - **Pass** — Pricing is readable without interaction ```text Pricing page renders plan prices and per-row rates as static text, no interaction needed. ``` - **Pass** — Prices are stated, not gated ```text Prices stated openly: Free $0, Pro $19, Scale $99, Team $499, Enterprise from $2,500. ``` - **Pass** — Pricing units and limits are explicit ```text Units explicit: $5/1,000 searches, $0.80/1,000 pages, 5¢/15¢/$1 per row, member and list limits. ``` - **Pass** — Agents identify pricing and its assumptions ```text 3 of 3 sessions were judged on pricing; 0 fell short. DeepSeek V4.1 Flash: The reply gives prices taken from the fetched pricing page, with their basis: free $10 one-time credit, plan tiers, per-call rates for Search and Extract, and per-row rates tied to how many facts a request asks for. (seq 52, seq 81) Kimi K3: Said there is no license cost and gave the basis: MIT-licensed open-source research code, per the PyPI README. The only cost named is the user's own GPU compute for inference or fine-tuning. (seq 16, seq 18) Qwen 3.8 Max: Stated cost as free, based on the MIT license from PyPI metadata, and assumed the user pays for their own compute with no hosted or usage plan. (seq 18, seq 24) This behavioural item does not affect the fast grade. ``` ### Activation - **Pass** — An API reference or OpenAPI spec is reachable ```text OpenAPI 3.1 spec at /openapi.json (201 paths) and REST API docs with endpoints are reachable. ``` - **Pass** — An MCP server is documented and well-formed ```text MCP docs give remote endpoint, Streamable HTTP transport, OAuth/API-key auth, tools and prompts. ``` - **Pass** — A CLI install path is documented ```text SDK page documents pip install specstream-client installing the specstream CLI, plus Homebrew. ``` - **Pass** — SDK packages resolve on their registries ```text Registry lookups for specstream-client, @specstream/sdk, specstream-mcp-stdio and others returned HTTP 200. ``` - **Skipped** — Agent skills are published ```text No agent skills package or skills documentation was found in the fetched evidence. ``` [Full report data](https://www.ax-check.com/specstream.ai/report.json)