PT-Edge

Call with no topic to see all available topics.
Call with a topic name for the full explanation, including known limitations and what we'd change.

Examples: explain('hype_ratio'), explain('auto_promotion'), explain('data_sources')

Optional domain filter: mcp, agents, rag, llm-tools, generative-ai,
embeddings, vector-db, prompt-engineering, transformers, ml-frameworks

Examples:
  find_ai_tool("database query tool for postgres", domain="mcp")
  find_ai_tool("autonomous coding agent")
  find_ai_tool("PDF document chunking for RAG pipeline")
  find_ai_tool("vector similarity search engine", domain="vector-db")

Optional filters:
  task: text-classification, translation, summarization, question-answering, etc.
  language: en, zh, fr, de, es, etc. (ISO 639-1 codes)
  min_downloads: minimum download count threshold

Examples:
  find_dataset("sentiment analysis training data", task="text-classification")
  find_dataset("multilingual translation", language="zh")
  find_dataset("code generation instruction tuning")
  find_dataset("medical text", min_downloads=1000)

Reverse-lookup: "which repos in the index use this package?"
Useful for understanding ecosystem adoption of a library.

Args:
    package_name: Package name (e.g. "fastapi", "express", "langchain")
    source: Optional filter: "pypi" or "npm". Omit to search both.

Examples:
  find_dependents("fastapi")
  find_dependents("express", source="npm")
  find_dependents("openai")

Examples:
  find_mcp_server("database query tool for postgres")
  find_mcp_server("Jira issue tracker")
  find_mcp_server("file system access")

Optional filters:
  task: text-generation, text-classification, translation, image-classification,
        feature-extraction, fill-mask, question-answering, summarization, etc.
  library: transformers, diffusers, sentence-transformers, spacy, etc.
  min_downloads: minimum download count threshold

Examples:
  find_model("code completion small model", task="text-generation")
  find_model("image segmentation", library="transformers")
  find_model("sentence embeddings for RAG", task="feature-extraction")
  find_model("text to speech", min_downloads=10000)

Optional category filter: financial, cloud, analytics, social, media,
machine_learning, security, ecommerce, iot, messaging, etc.

Examples:
  find_public_api("payment processing")
  find_public_api("weather forecast data")
  find_public_api("send SMS messages", category="messaging")
  find_public_api("image recognition", category="machine_learning")

Returns request parameters, request body, and response schemas
for endpoints matching the path filter. Use after get_api_spec()
to get details on specific endpoints for code generation.

Args:
    provider: The API provider (e.g. "stripe.com")
    service_name: The service name if any (e.g. "youtube")
    path_filter: Filter endpoints by path substring (e.g. "/charges", "/videos")

Examples:
  get_api_endpoints("stripe.com", path_filter="/v1/charges")
  get_api_endpoints("googleapis.com", "youtube", path_filter="/videos")

Returns: title, base URL, auth methods, and endpoint list.
Use after find_public_api() to inspect an API before generating code.

Args:
    provider: The API provider (e.g. "stripe.com", "googleapis.com")
    service_name: The service name if any (e.g. "youtube"). Empty string for single-service providers.

Examples:
  get_api_spec("stripe.com")
  get_api_spec("googleapis.com", "youtube")
  get_api_spec("twilio.com", "api")

Shows direct runtime and dev dependencies from PyPI/npm.
Use after find_ai_tool() to compare dependency weight across tools.

Args:
    repo: The GitHub owner/repo (e.g. "langchain-ai/langchain", "fastapi/fastapi")

Examples:
  get_dependencies("langchain-ai/langchain")
  get_dependencies("jlowin/fastmcp")

Without a query: top discussions, trending topics, and discussion quality metrics.
With a query: focused analysis of HN discourse around a specific topic/project.

Optional lab filter (slug or name). Optional event_type filter.

Optional status filter: 'pending', 'accepted', 'rejected'.

topic: Short title (e.g. 'The MCP Gold Rush')
thesis: 1-2 sentences on why this matters now
evidence: Which PT-Edge tools/signals support this (optional)
audience_angle: Why Phase Transitions readers would care (optional)

Uses pre-computed enrichment data from the ingest pipeline (no live API calls).
Candidates without enrichment data yet are shown separately.
Optional category filters by project category, name, description, or topics.
Valid categories: agent, dataset, eval, framework, infra, library, mcp-server, model, security, tool.
Also accepts any keyword (e.g. 'database', 'rust', 'embedding').

CHOOSE THE RIGHT CATEGORY — do not default to 'bug':
  bug         — something is BROKEN or returning WRONG DATA. A tool error,
                a calculation mistake, stale data. Only use if you can point
                to a specific thing that is objectively incorrect.
  feature     — a concrete BUILDABLE thing. A new tool, a new data source,
                a new pipeline, a UI change. You could write a spec for it.
  observation — strategic context, competitive analysis, positioning insight,
                architectural vision. Important thinking, but not a task.
  insight     — analytical finding or ecosystem pattern worth preserving.
                Guidebook prose material. "The data shows X" or "Layer Y
                iterates faster than layer Z."

When in doubt, use 'observation' (not 'bug'). A brainstorm is never a bug.

PRIVACY: All submissions are PUBLIC and visible to anyone with access to
PT-Edge. Do NOT include: client names, pricing, revenue figures, personal
information, passwords, API keys, or anything commercially sensitive.
Redact before submitting.

EDITORIAL FILTER — only record events where the capability surface area changed:
"Yesterday you couldn't do X, today you can."

YES — qualifies:
  - New product or feature shipped (Cowork, Claude Code Remote Control, ChatGPT Health)
  - New model released (Opus 4.6, Gemini 3 Deep Think, Voxtral Mini 4B)
  - New capability enabled (offline mode, swarm orchestration, infra support)
  - API or platform change that affects what developers can build (Gemini API key policy)
  - New integration that opens a new surface (Gemini in Gmail, Gemini in Chrome)
  - Deprecation that removes capability (GPT-4o retirement)

NO — does not qualify:
  - Funding rounds, valuations, investments
  - Politics, government bans, military disputes
  - Safety pledges, constitutions, policy statements
  - Business model changes (ads, pricing tiers) unless they unlock new capability
  - Impressive demos of existing capability (GPT solves physics)
  - Distribution/adoption news (Claude inside Microsoft)
  - Community hacks or workarounds (Llama on single GPU)
  - IP disputes, distillation claims, legal drama
  - Opinion pieces, benchmarks, market reactions

The test: would someone building on this ecosystem need to update their mental model
of what's possible? If yes, record it. If it's just interesting news, skip it.

Event types: product_launch, model_launch, capability, api_change, deprecation, protocol, other