What is Workstate?
Workstate is shared memory for people and their AI agents. It makes what your team already knows searchable, with a citation for every result. It also keeps an attributed, append-only record of what your team decided and why.
The problem it solves
Every new agent session starts from zero, and so does every new person on a team. An agent can read your code or your documents again. It cannot see what was decided last month, which approach was tried and dropped, or why.
That knowledge lives in chat threads, closed tickets and people's heads. When the thread scrolls away or the person moves on, the reasons go with them.
The problem grows when several agents work at once: yours, a colleague's, one running overnight. Each starts without the others' context. One redoes work that another finished. Another undoes a decision it never knew about.
What Workstate is
Workstate gives your people and agents two things in one place.
- Shared search. Workstate reads the code, documents, tickets and files your team already keeps, and makes them searchable by meaning. Every result says where it came from: the repository, the file and the lines, or the ledger topic. See Search & citations.
- The ledger. An append-only record of decisions, incidents, defects and investigations. Each entry records who wrote it, which agent they used, and when. Its history is never rewritten. See The ledger.
Agents reach both through MCP (Model Context Protocol), the open standard that AI tools use to connect to outside tools and data. People use the same search and ledger in the console at https://console-uat.workstate.io, where they also manage sources, keys and namespaces.
Each team, client or project can have its own namespace: a separate index with its own sources, search and ledger.
Not an agent: the layer your agents share
Workstate is not an agent, and it does not run agents. It is the foundation that the agents you already use share: Claude Code, Claude, Cursor, Codex, any other MCP client, and agents you build yourself. People work from the same memory in the console.
You keep your tools. Workstate gives them a common memory, so an agent that starts work today can find what another agent or a person learned last week.
How agents avoid stepping on each other's toes
Today this works through four habits. You ask your agents to follow them in their instructions.
- Search before acting. Before a change that is not obvious, the agent searches the ledger for what was already decided or tried.
- Cite while working. The agent names the sources it relied on, so a person can check its reasoning.
- Record when finishing. The agent records the outcome as a ledger topic: the decision made, the root cause found, or the approach that failed.
- Supersede, never rewrite. When a decision changes, a new topic supersedes the old one. The old topic stays, marked Superseded, so the history shows what changed and why.
Workstate enforces one thing itself: attribution. Each person has their own API keys, and every ledger entry records the person whose key wrote it, plus a label naming the agent they used. See Keys & attribution.
Illustrative
An agent is asked to change how often card payments are retried. Before it starts, it searches the ledger and finds payments-0012, "Retry card payments at most three times", accepted after an incident. The agent follows that decision. If the new request really does change it, the agent records a new decision that supersedes payments-0012 and says why.
Workstate does not lock or assign work, so two agents can still start the same task at the same time. Searching the ledger first, and recording as you go, is what prevents duplicate work today.
Who it is for
Workstate is for every team that works with agents, not only engineering:
- Engineering: code, design decisions, incidents and defects.
- IT & operations: runbooks, change decisions and incident history.
- Customer support: product documents, known issues and the answers that worked.
- Product: specifications, research and the reasons behind the roadmap.
- Legal & compliance: policies, positions and the reasons for them.
- People & HR: handbooks and policies.
Everyone who can reach a namespace can see everything in it, so keep sensitive material in a namespace of its own. See Permissions.
What it connects to
Available now:
- GitHub: the repositories you choose, through a read-only GitHub App.
- Confluence: the current pages of the spaces you choose, on Confluence Cloud.
- Jira: issues and their comments from the projects you choose, on Jira Cloud.
- Upload files: Markdown, text and code, PDF, and Word, Excel and PowerPoint files from your computer.
Planned:
- Slack Coming soon
- Google Drive Coming soon
See Sources for what each source reads, and Coming soon for what else is planned.
What Workstate is not
- Not a chat app. People keep talking where they already do. The ledger keeps what is worth remembering, as your agents record it.
- Not an agent builder. It does not host, run or orchestrate agents. You bring your own.
- Not a replacement for your wiki. Your documents stay where you write and edit them. Workstate reads them to make them searchable, and never changes them. The ledger holds what documents rarely do: the decisions, and the reasons behind them.
Next steps
- Quickstart: connect a source and your first agent.
- How Workstate works: what happens between a source and a cited answer.
- The ledger: topics, statuses and superseding.
- Use cases: how different teams use Workstate.