How It Works

    The investigation loop

    From live incident to reusable expertise.

    Each investigation starts from the user piloting their agent and ends by improving what the next agent can use.

    01 · Investigate

    Investigate using your existing agents.

    Expertise comes from how the engineer investigates: any facts, decisions, procedures, user steering, resulting outcomes, and more.

    02 · Capture

    Save the investigation.

    Prompt to save back to Tokonoma, make capture a standing instruction, or automate it at the end of the session.

    03 · Recall

    Recall the right context.

    Any connected cloud or local agent can recall the latest investigations, current evidence, ruled-out hypotheses, relevant procedures, and next steps instead of rebuilding new investigation from scratch.

    Recalled context

    Tokonoma· Investigation Expertise Recalled
    P2confirmed

    catalog-indexer writes are timing out. Restarts clear them, but they recur within ~40 mins

    investigated by Kevin Hou · 2026-07-07 · 5 hrs ·catalog-indexerpostgres
    Root cause
    The catalog-indexer's postgres connection pool config was too low for the steady-state traffic.
    Ruled out
    Slow queries because pg_stat_statements p99 was flat and normal
    The Fix
    Raised pool max_size from 50 to 100.
    References
    investigation 9c4ba62aby Kevin Hou

    Session receipt

    Tokonoma·Session Receipt

    Root cause: Checkout 5xxs came from connection-pool exhaustion

    This session reused 2 Tokonoma memories, drawing on prior investigation work by dana, alice and bob. Did not use 1 that was recalled but not relevant.

    investigationCheckout 5xx spike from connection-pool exhaustion

    Fix it gave you: bumped the pool ceiling and watched saturation

    by dana · 2026-04-12 · investigation 5b26d456

    procedureKafka consumer lag recovery

    Steps it gave you: scaled the consumer group, then drained the backlog

    updated 2026-05-19 · procedure 8a1f02c3

    Saved this session (Connection-pool exhaustion on checkout)
    04 · Maintain

    Automatically maintained.

    Expertise is automatically maintained: deduplicated, merged, re-ranked when irrelevant, and more. The shared expertise gets more useful as engineers and agents keep working with it.

    Explore the technology overview
    Install and connect

    Install once. Connect every agent.

    Tokonoma runs beside your workloads and connects through MCP, so local and cloud agents use the same investigation context without moving it outside your environment.

    01

    Deploy in your environment

    Install Tokonoma with Helm. Its MCP server, data store, and local models run in your Kubernetes environment, so investigation context stays under your control.

    02

    Connect every agent to the MCP

    Connect Claude Code, Cursor, Codex, or custom agents that run on your laptop or in the cloud to Tokonoma.

    03

    Run Tokonoma Onboard

    Import prior agent sessions, RCAs, skills, and runbooks so the first investigate starts with your team's proven expertise.