Dashboard

The dashboard is an operator view of the same durable state exposed by the versioned API. Access depends on membership state and role.

Overview

Overview is the organization’s work control surface. It leads with successful review, research, and delivery runs completed over the last seven UTC days, followed by current work counts, AI usage and cost, work that needs attention, active work, upcoming automations, and recent outcomes. Review-specific findings and quality metrics remain in Analytics.

Work items

Work items is the durable record for engineering requests received through GitHub, Slack, or Linear. A work item keeps its source conversation, ticket, selected repository, plans, artifacts, agent tasks, and delivery links together. Open a work item to inspect its current state or act on a plan that is waiting for approval.

Pull requests and findings

Pull requests lists synchronized pull requests with repository, state, latest review, and finding counts. Search or filter by open, closed, or merged state.

The pull request detail view combines:

• GitHub pull request identity and state;
• the latest and historical reviews;
• line-level findings and validation state;
• stored GitHub conversation;
• related workflows and agent tasks;
• finding feedback and publication outcome.

Finding records remain available even when a comment limit prevented every candidate from being published.

Review analytics and CSV export

Review analytics aggregates findings and feedback so operators can inspect volume, severity, acceptance, false positives, and other feedback reasons over time. Filters constrain the displayed interval or repository. CSV export downloads the filtered review data for offline analysis; it does not mutate reviews.

Automations

Automations lists scheduled questions, research, and builds created from Slack. Open a schedule to inspect its interval, next run, failure count, active occurrence, and delivery destination. Operators can pause, resume, stop, run now, or change the interval without losing occurrence history.

Activity

Activity contains the diagnostic records behind user-facing work. Its Agent tasks and Workflow runs tabs are useful when an operator needs to inspect execution details or troubleshoot a failure.

Agent tasks

Agent tasks shows the unit leased by a worker. The detail page records:

• provider, model, and reasoning effort;
• sandbox provider and identifier;
• attempts, timing, usage, and status;
• input snapshots and instructions;
• produced artifacts and rendered output;
• sanitized failure details.

A retry appears as attempt history. Use task data to diagnose clone, sandbox, tool, provider, artifact, and timeout failures.

Workflow runs and events

Workflow runs is the end-to-end state machine for pull request reviews, ticket plans, implementations, channel questions and research, mentioned change requests, recurring occurrences, and description generation.

The run detail page includes ordered persisted events. Where available, the event stream provides live progress while retaining the same events for later inspection. Correlate delivery, workflow, job, task, attempt, sandbox, and publication identifiers with structured server and worker logs.

Repositories

Repositories synchronized from the GitHub installation appear here. New repositories start paused. Admins and super users can:

• enable or pause workflows;
• set review triggers, draft behavior, exclusions, strictness, and limits;
• set implementation pull requests to draft or ready;
• enable description regeneration and select a discovered template;
• choose category-specific runtime overrides;
• manage repository-scoped secrets;
• choose approved read-only secondary repositories.

The repository detail view reports whether each effective setting comes from the instance or repository.

Learnings

Learnings shows proposed reusable guidance with actor, repository, interaction, and approval provenance. Admins and super users can approve, enable, or disable a learning. A proposal is never active merely because an agent produced it.

Integrations

Integrations is where operators connect and inspect:

• the GitHub App used for repositories and trusted publication;
• Slack workspaces used for requests, approvals, results, and optional alerts;
• Linear workspaces used as ticket sources;
• private artifact storage for HTML research reports;
• remote MCP servers, agent profiles, sandbox profiles, and skills.

Integration credentials are not returned to the browser after storage.

Settings

Settings are divided into:

• General — organization identity, ticket defaults, and the OpenAI-compatible conversation model configuration.
• Members — role and active-state management plus membership-change history.
• Agent — Codex or Claude provider, model, and default reasoning effort.
• Sandboxes — Docker, Daytona, or Lambda provider fields.
• MCP servers — remote HTTPS HTTP or SSE transports and encrypted header references.
• Skills — packaged agent instructions.
• Instructions — shared custom instruction text.
• GitHub — App identity and super-user reconfiguration.
• Alerts — opt-in Slack operational alerts and their selected destination.

Member management

USER can view product state. ADMIN can also manage repositories, runtime configuration, secrets, and learnings. SUPER_USER can manage the GitHub App and users.

Disabling a member revokes all active sessions. Tasuku supports several super users but prevents disabling or demoting the last active one. See Members and access.