Logs, metrics, traces
Any existing provider
Deployment, access and security.
Keep your telemetry providers. Run Tokonoma in your cloud, connect your coding agent, and choose whether to enable Tokonoma’s Agent-first Telemetry Platform.
See where Tokonoma fits.
Connected tools + APIs
Queries and evidence
Your cloud account
Customer Kubernetes
Tokonoma
Agent endpoint + investigation knowledge
Monitoring and unattended runner when enabled
Log + trace storage
Log + trace storage
Your cloud object storage
Optional
Model Context Protocol
Tools and findings
Claude Code, Codex, Cursor
Work with the evidence in your repository
Your agent’s model connection
Tokonoma’s model-backed features
Context may leave your cloud
Model processing
Processes the context sent by enabled features and your connected agent.
Choose the components you need.
Tokonoma core
- Agent endpoint
- Investigation knowledge
- Existing telemetry providers
Agent-first Telemetry Platform
- Log + trace indexing
- Your cloud object storage
- Alongside existing tools
Deployment options and existing telemetry
Install Tokonoma with a Helm chart in a cluster you manage, or use a Tokonoma-operated AWS deployment in your account.
The core installation stores investigation knowledge and provides the Model Context Protocol (MCP) endpoint that agents connect to. Tokonoma can use any existing telemetry provider through connected tools and APIs. Its own telemetry platform is optional.
Understand the data and model boundary.
Customer context
Findings, records, messagesUnattended investigationsClaude
Memory processing + Slack triageAnthropicOff by default
Recurrence decisionsTypeSafe’s JevOptional
Model-backed features can send context outside your cloud.
Which content reaches model providers?
Unattended investigations use Claude. Your connected coding agent uses the returned evidence in its own model session. Where telemetry is stored and where model processing happens are separate deployment decisions.
When server-side memory processing is enabled, it sends record titles, summaries, bodies and reference URLs to Anthropic. Slack triage also sends the message text it processes. This server-side processing is off by default.
If TypeSafe’s Jev is enabled, detection context and candidate investigation records are sent to TypeSafe for recurrence decisions.
Configure access for each connection.
People + agents
Configured sign-inTokonoma access
Deployment policyConnected tools
Permissions per integrationIdentity and local access
Deployment owners configure authentication and network exposure. Remote MCP connections can require authentication through AWS Cognito, Google Identity Platform, or direct Google Workspace or Okta SAML.
The local quickstart uses a Kubernetes port forward and starts without application-level authentication. Enable authentication when access should require sign-in.
Customer role and connected-tool permissions
For a Tokonoma-operated AWS deployment, the customer creates a cross-account role used to provision and operate the installation.
The optional Grafana adapter uses a Viewer service account restricted to the required data sources. It exposes selected read-only tools for metrics, logs, traces and profiling, with dashboard editing and other write tools excluded. Review permissions separately for other integrations and coding agents.
Verify published releases.
Signed image
Sigstore cosignSigned software bill of materials
Bound to the image digestScanned daily
Newly disclosed vulnerabilitiesRelease verification and deployment security
Published Tokonoma server images are signed with Sigstore cosign and carry a signed software bill of materials. Verification checks bind the artifact to the release workflow and image digest.
Published images are scanned daily for newly disclosed vulnerabilities. Deployment configuration, network policy and identity-provider settings remain part of securing your installation.
Deployment questions.
Can Tokonoma use our existing telemetry?
Yes. Tokonoma can use any existing telemetry provider through connected tools and APIs. Tokonoma’s own telemetry platform is optional.
Does storing telemetry in our cloud keep all processing there?
Model-backed features can send context to external providers. Unattended investigations use Claude, and optional server-side memory processing sends record content and Slack triage text to Anthropic. Review those model connections as part of deployment setup.