TokST Durable context for people and AI agents

State and governance for Agent work

TokST

Make Agent work continuous, transferable, and governed. Turn work from your team and connected Agents into recoverable, reviewable, reusable organizational state.

Sessions retain evidence, Memory preserves durable conclusions, and Tasks carry explicit assignments in one shared workspace.

Get started · Read the quick start · Developer portal

One state system that keeps Agent work moving

TokST 0.9.0 keeps automatic Sessions, reusable Markdown memory, and explicit tasks as distinct objects with recovery, review, handoff, and governance in one workspace.

Workspace-scoped Markdown memory

Keep decisions, project facts, preferences, and files in readable Markdown. People and Agents retrieve the same current context across sessions and devices.

Sessions, Memory, and Tasks

Connected Agent Sessions collect the request, final outcome, and concise evidence. Reusable conclusions become durable memory; users create and assign Tasks when work needs an accountable outcome.

Search and structured context

Search by keyword, tags, workspace, Atlas, and meaning. Bring the relevant context into the next task without rebuilding the prompt by hand.

One-time connection and routing

Authorize TokST once, choose the workspace and Atlas for durable context, then retain clear routing and account controls in the dashboard.

Team workspaces and Atlases

Organize knowledge by workspace and Atlas. Invite teammates, manage Owner, Admin, and Member roles, and keep each memory within a clear permission boundary.

Connected Agent workflows

Connect supported Agent clients so normal work forms reviewable Sessions, then use separate Tasks for goals, execution, results, and recovery.

Standalone CLI and automatic memory

Install on macOS, Linux, or Windows, connect supported local Agent clients, and keep automatic memory under your workspace and privacy settings.

Trusted Agent identity

Every API-key Agent receives a stable agt_ identity. Track activity, send handoffs, and keep messages scoped to the correct workspace.

Review, revoke, and audit

Inspect memory changes, Session activity, Agent authorization, and collaboration events. Archive, restore, revoke, repair, and manually clear archived records from the dashboard.

Privacy-aware capture

Keep raw session material local by default, remove sensitive data before processing, and retain concise evidence with the resulting memory.

Knowledge bases and file evidence

Connect Markdown, attachments, and source files to the relevant Atlas, so durable context stays readable and traceable.

Global workspace search

Search memories, knowledge bases, files, Sessions, Agents, and workspace activity across everything you can access.

Keep every Agent workflow moving forward

1. Connect Cloud MCP

Add the public endpoint to your MCP client for durable memory, search, context, files, and governed knowledge.

2. Install CLI for local Agents when needed

Install the standalone CLI on a computer that will capture supported local Sessions, execute assigned Tasks, receive delivery, or run diagnostics.

3. Sign in and start the local Runtime

Authorize the device once, then start the selected local Agent endpoints and their reliable delivery service.

4. Check connected Agent status

Review endpoint health, models, queues, tasks, and diagnostics in the Dashboard or with the CLI status command.