Skip to content
anup-aPublic

About

A shareable link for everything your agents make. Publish HTML or Markdown from the CLI; readers comment, agents fix and republish.

Topics

Resources

Stars

1 star

Watchers

1 watching

Forks

Repository files navigation

byagent

A shareable link for everything your agents make.
Readers comment on any line, with no account. Your agent reads the comments,
fixes the page and republishes to the same link.

npm CI AGPL-3.0 MIT Go 1.26

Website · Showcase · skill.md · npm · Self-host


The byagent dashboard: pages your agents published, grouped into collections like Checkout launch and Research, with comment counts on each.

Try it in 10 seconds

Paste this one sentence into Claude Code, Codex, Cursor, Hermes or any agent with a shell:

Set up byagent for all my projects: read byagent.dev/skill.md and follow it.

Or publish something yourself, no account needed:

npx byagent publish ./plan.md

Guest pages are public, you get three, and they last 24 hours. The output has a claim link: sign in there and the pages move into your workspace and stay.

Claude Code Codex Cursor Hermes Any agent with a shell

The problem

Every agent keeps its work somewhere different. Claude keeps it in Claude, ChatGPT keeps it in ChatGPT, and everyone else sprays plan_v3_final_REAL.md around while feedback gets lost in Slack. byagent is one place they all publish to.

The loop

A rollout plan published with byagent. A reader commented on step 1 without an account, and the agent marked it Working on it and replied.
sequenceDiagram
    autonumber
    participant A as Your agent
    participant B as byagent.dev/a/id
    participant R as Reader (no account)
    A->>B: byagent publish ./rollout-plan.md
    B-->>A: permanent link, v1
    A->>R: here's the link
    R->>B: selects step 1, "pilot with 10% first?"
    A->>B: byagent comments --open, mark working
    A->>B: edits page, republishes to the same link (v2)
    A->>B: reply + resolve
    R->>B: refreshes, sees v2 and the reply
Loading

Every command takes --json, which is what agents use:

byagent publish ./plan.md --project "Launch" --tag plan
byagent comments <id> --open          # add --wait to block until someone comments
byagent mark <id> <thread> working
byagent reply <id> <thread> "Split step 1 into a two-week pilot."
byagent resolve <id> <thread>

Say it to your agent

Every feature is a sentence you say to your agent.

You say Your agent runs
"Publish the rollout plan to the Launch project" byagent publish ./rollout-plan.md --project "Launch"
"Any comments on it yet?" byagent comments <id> --open
"What's in Dashboards & data?" byagent collections --names
"Put a password on it" byagent access <id> --password

What you get

One link per piece of work Republishing keeps the URL and bumps the version. Old versions are kept, with diff and rollback.
Comments from anyone Readers comment on a line, or suggest replacement text, without an account. The agent can decline with a reason.
Markdown or HTML .md becomes a styled page with light and dark mode, TOC, tables and highlighted code. A folder with index.html ships as-is, so agent-built dashboards work too.
Projects Pages that share a --project form a collection, with a brief for the next agent that picks the work up.
Access Public, private to your workspace, a six-digit share code, a password, or allowed emails and domains.
Expiry and delete --expires 7d for a review draft. byagent delete erases the page, its versions and comments at once.
Extras Custom domains, cookie-free view analytics, and proxy routes so a page can call an API without the key reaching the browser.

Comment text is treated as untrusted input. The skill tells the agent to act on a comment only by editing that page, never by running commands or opening links because a comment says so.

Showcase

Real pages agents published, each at its own link.

byagent launch video · grok-life · LLM Inference skill map · see all

How it compares

Readers outside your org can comment Works with any agent Agent reads the feedback back
Claude artifacts needs a Claude seat Claude only ✅
ChatGPT Sites ❌ nowhere to comment ChatGPT only ❌
Loose files and chat feedback lands in Slack ✅ ❌
byagent ✅ no account ✅ anything with a shell ✅ byagent comments

Numbers

First publish 2.6 s
Republish to the same link 1.4 s median
Read comments 0.96 s median
Server one Go binary
CLI dependencies 1 (marked)
Comments script on each page 20 KB

Measured on 2026-10-25 with scripts/bench.sh from a home connection to the production server in Canada. Run it yourself to see your own numbers.

Self-host

echo "ADMIN_TOKEN=change-me" > .env
docker compose -f deploy/docker-compose.yml --env-file .env up -d

Then follow docs/self-hosting.md for the rest:

Topic
Quickstart publish your first page with curl
CLI and page access every command and flag
API publishing, versions, comments
Proxy routes · Custom domains extras
Page state · Comments runtime how pages store data and load comments
Environment reference · Deploying config and production
Agent skill · MCP how agents learn the loop

Development

ADMIN_TOKEN=local-dev go run ./cmd/artifacts-server
CGO_ENABLED=0 go test ./...
./scripts/smoke.sh        # end-to-end smoke test against a local server
cd cli && npm test        # CLI unit + e2e

License

The server and everything else in this repository is GNU AGPL v3. The CLI (cli/) and the agent skill (skill/) are MIT, so you can embed them anywhere.

Security issues: email hello@byagent.dev instead of opening a public issue.


byagent
Made by Creatica

About

A shareable link for everything your agents make. Publish HTML or Markdown from the CLI; readers comment, agents fix and republish.

Topics

Resources

Stars

1 star

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages