03 · The Harness

Any harness, any model.

Known is one memory system for your entire fleet of agents and the team that runs them. Whatever harness an agent runs in, and whatever model is behind it, it reads the same record before it works and writes what it decided back under the name of the person who ran it. Your coding agent knows your code; Known tells it what your company knows. Nothing is filed by hand, and nothing is trapped in one tool.

  • Claude Code
  • Codex
  • Grok Build
  • Pi
  • any agent that runs a CLI
brock@acme — claudeknown · acme · connected
coding agent · ~/acme/billing

Kn known connected · org acme · 41 people · 2 agents active

wiki handbook · 65 pages · signed in as brock

before you touch billing: what did we decide about annual refunds?

Bash(known acme wiki get decisions.annual-billing)revision 4 · Dana · 2d ago · on file (expand)
Annual plans bill up front. Refund window: not yet decided.
1 note open · Sam: do we prorate?

Not settled. Annual plans bill up front and the refund window is still open, with Sam asking whether we prorate. Source: decisions.annual-billing, revision 4.

finance settled on 30 days this morning. is anyone else on this?

Bash(known acme wiki search "annual refund")3 pages · people/dana · decisions/annual-billing · runbooks/invoice-template (expand)
people/dana · owns billing policy · changed 1d ago by pi, run by Priya
decisions/annual-billing · Dana · 2d ago
runbooks/invoice-template · Priya · 5d ago

Dana owns billing policy, and Priya’s agent touched her page yesterday. I’ll write the decision as its own page and point both of theirs at it.

write it up so their agents see it too

Bash(known acme wiki set decisions.annual-billing-refunds --file notes.md)revision 1 · written · brock · agent mark pi · +18 lines · note closed

next session, Priya’s agent reads this page before it touches billing.

Try “what changed in billing this week?”

? for shortcuts · click a ⏺ line to expand its output

1

Agents read it first

Before an agent touches billing, it reads what the company decided about billing. One CLI, any harness, as you.

2

Agents write it back

What gets decided in a session lands as a page, attributed to the person who ran the agent, with the agent's mark.

3

The team keeps it honest

History, diffs, and notes on every page. The change log says what is new since you last looked.

One memory system

Use whichever harness and model suits the job. The knowledge stays in one place.

Known is not a plugin for one agent. The known CLI is the whole integration, so any harness that can run a shell command reads the record and writes to it, whatever model is behind it. A person can run Claude Code for one task and Codex for the next. Two teammates can each pick their own. Their agents read the same pages, write under their own person's name, and leave a mark saying which harness did the writing. Nothing about the company's knowledge depends on which tool, or which model, someone chose this morning.

Claude CodeSupported

Anthropic's terminal agent. It calls the known CLI through its Bash tool; the session header shows the connection.

CodexSupported

OpenAI's coding agent. Same CLI, same login, same pages; a decision it writes is read next by whichever agent runs after it.

Grok BuildSupported

xAI's coding agent. Runs known from its terminal as the person who signed in, with their organization's permissions.

PiSupported

The pi coding agent. Small and terminal-first; it shells out to known like any other tool it runs.

any harness · one recordShared
  • one loginSign in once per organization. Every harness on the machine uses that session and speaks as youTravels
  • modelClaude, GPT, Grok, or an open model. Known reads the same and writes the same; the mark records which agent it wasAny
  • same pagesTwo agents in two harnesses read and write the same wiki; neither keeps a private copyShared
  • agent markEach write names the person and the harness that made it, so a diff says who and whatRecorded
  • hand-offWhat one agent writes at the end of a session, the next agent reads at the start of its own, whatever it runs inKept
The harness and the model are the person's choice, job by job. The record is the company's.

Why the harness needs it

What an agent learns today stays where it learned it.

Per-tool memory

Each harness keeps its own notes for one person on one machine. A teammate's agent cannot read them, and neither can the teammate.

Files in the repo

Context files travel with one repo and one branch. Nobody knows who wrote a line, when it went stale, or which agent last changed it.

Chat transcripts

Decisions get made inside a session and disappear with it. The next session, and the next person, work them out again from scratch.

Known keeps what was decided outside any one tool, repo, or session, where every person and every agent on the team can read it, write to it, and see who changed what.

harness · what the agent getsConnected
  • identityThe agent runs as the person who signed in, with that person's organization and permissionsScoped
  • readGet, search, tree, and history on the wiki; every answer names the page and revision it rests onCited
  • writePages the agent writes carry the person's name and the agent's mark, and open as diffs for the teamAttributed
  • harnessClaude CodeCodexGrok BuildPi· any agent that runs a CLICompatible
One CLI carries the person's identity and the company's record into any agent session.

Get started

See it with your own record.

Pricing