AI Builder
Pipe, our open-source fork of the Pi harness, turns a conversation into a working data pipeline you can inspect, edit, and run.
Build stable pipelines to public data
Describe the site you need. Pipechat builds and validates reliable extraction rules against real captured pages, then runs them deterministically to structure, enrich, and deliver consistent web data to your database or agents.
Sources and Recipes
Pick from a growing list of maintained sources and a library of premade pipeline recipes — run them as they are, or tweak them to fit your setup.
Model Agnostic
Connect the models that fit the job — from paid LLM providers to Hugging Face, Groq, OpenRouter, and self-hosted open-source models.
Self Healing
LLMs monitor and repair external connections when sites change — whether you use a Pipechat-maintained source or connect your own source from the open internet.
Cloud or Local
Use hassle-free managed cloud compute for scheduled runs, or run the same pipeline locally on your own machine and infrastructure.
MCP Servers
Host your data behind an MCP server and query it directly from AI chats and agents such as ChatGPT and Claude.
PipelineCode FilesRun History
⌗ − 100% ＋
Ask Pipechat to build or change this pipeline…
SourceAddressSource typeAvailabilityPipeline
Federal Register
https://federalregister.gov/documents
Government
Available
In pipeline
→
Companies House
https://find-and-update.company-information.service.gov.uk
Government
Available
Not linked
→
FDA Drug Safety
https://fda.gov/drugs/drug-safety
Government
Available
Not linked
→
UK Legislation
https://legislation.gov.uk
Government
Available
Not linked
→
EUR-Lex
https://eur-lex.europa.eu
Government
Available
Not linked
→
SAM.gov Opportunities
https://sam.gov/content/opportunities
Government
Available
Not linked
→
Apple Newsroom
https://apple.com/newsroom
Corporate
Available
Not linked
→
Microsoft Source
https://news.microsoft.com/source
Corporate
Available
Not linked
→
NVIDIA Newsroom
https://nvidianews.nvidia.com
Corporate
Available
Not linked
→
Shell Media Releases
https://shell.com/news-and-insights/newsroom
Corporate
Available
Not linked
→
BHP News
https://bhp.com/news
Corporate
Available
Not linked
→
Rio Tinto Releases
https://riotinto.com/en/news/releases
Corporate
Available
Not linked
→
SOURCES AND RECIPES
Build with AI, or start from our maintained sources and recipes.
The Source Library holds sources we keep current for you — address, extraction rules, validation evidence, and the full version history. Connect one to a pipeline and it behaves like any other block, except you never wrote it and you never have to fix it. Recipes go a step further: prebuilt pipelines wired end to end that you can run as they are, or tweak into your own.
- Monitored on a schedule
- Fixed as a new version
- Reusable across pipelines
- Recipes you can run or tweak
ProviderTypeModels / service
Anthropic
Model
Claude Opus · Sonnet · Haiku
→
OpenAI
Model
GPT · o-series · embeddings
→
Google
Model
Gemini Pro · Flash · Gemma
→
Hugging Face
Model
Open models · Inference API
→
Groq
Model
Llama · Qwen · Mixtral
→
OpenRouter
Model
One API · hundreds of models
→
PostgreSQL
Database
Tables · JSONB · pgvector
→
Supabase
Database
Managed Postgres · Storage
→
BigQuery
Warehouse
Datasets · warehouse tables
→
MongoDB
Database
Documents · Atlas Vector Search
→
Snowflake
Warehouse
Warehouses · stages · streams
→
Amazon S3
Storage
Parquet · JSONL · CSV
→
MODEL AND STORAGE AGNOSTIC
Any model. Any database.
Pipechat is the layer between the two, not a replacement for either. The model that reads a page and the store that keeps the records are both settings — change one and the same pipeline keeps running, on your keys and inside your own infrastructure.
- Any model providerAnthropic, OpenAI, Google, or a local model behind your own endpoint — set per pipeline or per block.
- Any storage systemPostgres, Supabase, BigQuery, MongoDB, object storage, or a write straight to your own API.
- Swap without a rebuildThe generated code reads the provider and destination from config, so switching costs a setting, not a rewrite.
LOCAL OR CLOUD
Run the same pipeline wherever you want.
Use Pipechat Cloud when you want managed compute and scheduled runs, execute locally inside the desktop app, or download the complete codebase and operate it yourself outside Pipechat.
- 01Run in Pipechat CloudManaged compute, schedules, logs, and retries without maintaining infrastructure.
- 02Run in the desktop appKeep execution and data on your own machine with the same visual workflow.
- 03Download and self-hostExport the full Python project and run it in any environment you control.
block.pyrun_pipeline.py
# The same project runs in Cloud, Desktop, or your own environment. from __future__ import annotations import os from pipeline_runtime.context import as_records from pipeline_runtime.models import complete # Provider keys and runtime target stay in environment config. API_KEY = os.environ["ANTHROPIC_API_KEY"] RUNTIME = os.getenv("PIPECHAT_RUNTIME", "local") def run(payload, context): records = as_records(payload) for record in records: record["industry"] = complete( api_key=API_KEY, model="claude-sonnet-5", prompt=f"Classify this company: {record['body'][:2000]}", ) return records