Mira Admin

Users

Pending invites

Demo Testers

Demo pathways are windowed logins into the shared demonstration profile — each tester can only sign in and run Mira during their scheduled window (a weekday plus a start–end hour in their own time zone), and only between the pathway's valid-from and valid-until dates. Generate a staggered grid of ten, or add pathways one at a time; edit each window and the per-tester message inline, then send each tester a branded Mira email with their credentials, window, and message. Passwords are shown only when a pathway is created, seeded, or reset.

Add a demo pathway

Demo pathways

Usage Dashboard

Daily Usage (Last 30 Days)

By Model

Atelier Analytics

Completed runs only. Token/cost/turns are rolled up from per-call events at run completion; token coverage shows the share of runs with any token data logged (Claw-subprocess and dev-worker calls currently under-report tokens, so low coverage means the average understates real burn).

Team API-Key Bundles

Create shared-key bundles (e.g. "Alpha testers") and grant them to users so collaborators can use Mira's LLM features without supplying their own API keys. Users' personal keys always take precedence over team keys when both are available.

Per-bundle Spend

Storage Allowance

Every user gets a slice of disk space for their library (the PDFs and files they upload or import). When a user fills their allowance, new uploads are blocked until they delete files or you give them more room here. Usage is measured directly from each user's stored files, so the numbers are always exact.

Tip: set a user's allowance to 0 for unlimited (no cap). Leave a user's box blank to use the default. "Split pool equally" sets the default to the pool divided by the number of current users.

Per-user allowance

System Health

Backups

Crash Reports

Configuration

Activity Log

Pipeline Failures & Proposed Fixes

Cross-user log of pipeline-stage failures where automated rescue did not succeed, with the root cause and the fix proposed by the self-healing subagent. Proposals are logged only — none are applied automatically.

Library Requests

Users who have asked their university library to wire its digital holdings into Mira via the public bulk-ingest API. Reach out to the user (and offer to liaise with their library's IT desk) to close the loop. Click "Acknowledge" once you have contacted them.

Researcher-collaborator applications

Applications submitted from the public /apply.html page. Approve grants the applicant full access (the onboarding waiver) and, if they have no account yet, generates one and emails them a set-password link. Decline just records the decision.

Server Log

Tail of the structured server log (and the launcher log). Newest entries first. Use this instead of a shell — no SSH needed to see why the server crashed.

Tier Stats

Per-tier metadata-extraction success rate across all library files. Each row shows how many files were processed by that tier, and what fraction produced bibliographic data (success), found no match, had no input to query, or errored. Use this to find tiers that routinely cost time without producing results — candidates for later cost-cutting.

Diagnostics

Run intake-parse + GROBID against a specific library file and inspect the resulting tier ledger, persisted metadata, and raw artifacts. GROBID fulltext can take 30-180 seconds on large PDFs.

Recent diagnostic runs

Pill Matrix

Pipeline Health

Self-Heal Queue

Intake Trace — live pipeline diagnostic

Subscribes to the live pipeline event stream. Run a file (Acquisitions / Library) and watch which stages and sub-steps actually fire, when, how long they take, and exactly where the chain stops. States are derived directly from events — not fired, running, ok, failed, and not reached are distinct, so a dark stage is never ambiguous. Read-only observer; it does not alter the existing progress pills.

No files seen yet on this connection. Run a file to populate the trace.

Model Strength & Stance Weights

Each model has a 0–1 strength score. In claim adjudication the strong planner works ~90% off the evidence; a subagent's stance is weighted by its strength vs the planner (0.1…0.5). Edit a score to override it (persists to data/model-strength.json, picked up on the next run); clear the box + Save to revert to the built-in/pattern default. Add a model id to assign a score to one that isn't listed.