The research harness
The Knowledge pack post on hraness.com explains how Sponge works with ALGAL, Excalibur (xcb), Wordcell, and GhostGet.
Read the Knowledge pack postDeep-research harness
Research that runs on your machine and cites its sources.
Sponge runs as a background service with a terminal chat and a terminal status view. Ask a question, and it gathers sources, keeps their text, and writes a report that cites them.
In the terminal chat: /deep What do field studies say about coral reef recovery after bleaching?
Sponge is in private preview, and there is no public install yet. You can use Sponge Library, the hosted library at sponge.computer, today.
spongev2 project create ~/Research/reefs --name Reefs \
--question 'What does the evidence say about reef recovery?'
spongev2 tui ~/Research/reefsCommands from the private preview. Read the Knowledge pack post, or call the Sponge Library API.
- Research
- Deep mode plans the research, gathers public sources, recalls project memory, and checks a draft before it updates the report.
- Improve
- ALGAL tests a new brief-research strategy against the current one. It takes effect only when you adopt it, and you can restore the previous one.
- Publish
- Wordcell stores and searches your documents, and publishes a static edition you host or selected documents on wordcell.io.
Four Hraness tools do the work
Sponge brings four Hraness tools together, and each has one job.
- ALGAL
- Keeps project memory, runs research tasks, and scores a proposed brief-research strategy against the current one on a fixed set of research cases. You adopt a new strategy or restore the previous one.
- Excalibur (xcb)
- Handles every model call through the coding subscriptions you already have. Your provider sign-ins stay in xcb.
- Wordcell
- Keeps documents and search, and publishes static editions or pages on wordcell.io.
- GhostGet
- Captures pages you are signed in to. Sponge records only which login to use, never cookies or credentials.
From a question to a cited report
Create a project with a question. Sponge gathers the sources and keeps the report current.
Create a project
Give it a question and a folder. Sponge saves each source’s text and records its work, so a task continues after you close the terminal.
Ask, or go deep
Ask in the terminal chat, or use /deep for a planned investigation within the limits you set. Missing search results and partial evidence stay visible.
Keep it current
Schedules refresh the report from saved sources, and deep schedules can gather new public sources. Your own notes outside the generated section stay as you wrote them.
Where to find Sponge today
You can read the daily paper digest Sponge writes and use the hosted library today.
Daily papers
The Hraness literature pipeline runs Sponge every day to watch for new papers, write a digest, and publish it at hraness.com/papers.
See the papers
Sponge Library
Save articles and PDFs as dated, readable copies with your highlights and notes, and let your own AI agent, such as Codex or Claude Code, read and cite them through the API.
Open Sponge Library
You decide what changes
Sponge researches and checks its drafts. You decide which strategy it uses and what it publishes.
- Runs on your machine
- Projects, source text, and reports live in folders you choose. Sponge has no model account of its own.
- Strategy changes wait for you
- A new brief-research strategy applies only after you adopt it, and you can restore the previous one. Deep research is not evolved.
- Citations you can check
- Each report cites saved sources. The citation check confirms that each reference points to a saved source, not that every claim is true.
Before you begin.
What is Sponge?
A deep-research harness that runs on your machine as a background service, with a terminal chat and a terminal status view. It saves the text of the sources it uses and writes reports that cite them.
How do I get it?
Sponge is in private preview, and there is no public install yet. The Knowledge pack post on hraness.com describes it and the tools it is built on.
Which models does it use?
Sponge sends every model call through Excalibur (xcb), which uses the coding subscriptions you already have. Your provider sign-ins stay in xcb.
How does Sponge improve its research?
ALGAL proposes a new strategy for brief research and scores it against the current one on a fixed set of research cases. Nothing changes until you adopt it, and you can restore the previous strategy.
Can it read pages I sign in to?
Yes, through GhostGet, which handles signed-in capture and paces requests for each login. Sponge records only which login to use, never cookies or credentials.
Who uses it today?
The Hraness literature pipeline runs Sponge every day to watch for new papers, write a digest, and publish it at hraness.com/papers.
What is Sponge Library?
The hosted library at sponge.computer, and the first Sponge product. It keeps a dated copy of each article or PDF you save, anchors your highlights and notes to it, and gives your agent a scoped, expiring token to read and cite it. Sign in with your Sponge account, or create one.
Who makes Sponge?
Hraness builds Sponge. Hraness is a software studio that makes tools for AI agents and for people, from agent memory to music. Most are open source.
Private preview
Deep research that runs on your machine and cites its sources.
Read how Sponge fits with the rest of the Knowledge pack, or save your reading in Sponge Library today.
Sponge is in private preview, and there is no public install yet. You can use Sponge Library, the hosted library at sponge.computer, today.