Press Start to capture from your microphone. Tick “Include system audio” to also capture audio from a video call or media playback.
Stop voice mode to use Meeting capture.
Listening…
Knowledge Graph
Related Context
No local matches found.
LLM Answer
Projects
No watched items
Loading issues...
No open issues found
Insights
Loading insights...
Select an insight from the sidebar or create a new one.
No meeting profiles yet. Run /recap to auto-create one, or /meeting-profile [name] to build one manually.
No user model yet. Run /usermodel in chat to build one.
Connect your Slack workspace to use Slack Alerts.
Alerts
#confirmedrejected
Matched:
No matched alerts yet.
Create an alert to start monitoring, or select messages from the feed.
Alert Definitions
Distilled rules:
Feedback: |
Created:
No alert definitions yet. Create one above or select messages from the feed.
Feed
#
Messages from monitored channels will appear here as they arrive.
Create Alert
Knowledge Graph
These channels will be skipped when reading Slack during Build. Use this to exclude noisy/automated channels.
#
Connect Slack to configure excluded channels.
Loading channels...
TARS automatically removes Knowledge Graph and warehouse data older than this window on startup. Set to 0 to keep everything forever.
Applies the limit immediately (otherwise it runs on next launch).
Loading graph...
Entities
Relationships
Predicate Types
Warehouse Chunks
Embeddings running on CPU / re-indexing — retrieval may be degraded.
Graph
● Top entity● Entity━
Warehouse
All●
Avg Quality
Embedded
Entity-Linked
Summaries
Entity: Source: Clear
Relationships by Type
Top Entities
Last updated:
Press Build to populate it from your Slack conversations, Jira tickets, and calendar events. This powers /people, /prep attendee context, and semantic search.
Worker LLM Log
Purge Knowledge Graph
This will permanently delete all entities and relationships. This cannot be undone.
Purge Warehouse
This will permanently delete all semantic chunks. This cannot be undone.
No projects yet. Create one to start tracking work and journal entries.
Project Graph
Could not resolve person "" — showing the project only.
Nodes
Links
Journal Entries
Graph
●
Graph visualization is desktop-only. Use the inspector below to browse nodes.
Timeline
⚠ item(s) without dates — not shown on the timeline
Timeline is desktop-only — use Report.
No report available — select a project.
items ·
blocked
items
· blocked ·
Blocked:
Unassigned
items
· blocked
Blocked:
External Blockers
←
Recent Decisions
· ·
Last journal entry:
· · created by
Links
Journal
—
Add Node
Link Nodes
Add Journal Entry
New Activity
New Project
No situations tracked yet. The Hunch Engine opens one when it starts investigating an open question.
p= —
observations ·
hypotheses
· entropy
· last activity
· opened
Hypotheses
p=
()
Surfaced Summary
Observations ()
·
(reliability ) ·
cluster ()
Belief History
():
, =
Calibration: resolved prior claim(s) on record.
Scenario Moves
Scenario is now the accepted plan. Draft scenarios will build on top of this.
JIRA ACTIONS NEEDED:
No Jira updates needed for this scenario.
Defer until:
Set leave:
Remove person:
Create a scenario to defer projects
Reassign to:
Add co-assignee:
Block until:
Adjust estimate (days):
Create a scenario to reassign tasks
Required Skills:
Workforce & Skills
Import from Jira or add people manually.
No change
Chats
▶
▶Archived
Loading…
No archived chats
Q Light
Activity
No scans yet.
···
→
Setup tips
💡 Open the KG tab and press Build to populate your knowledge graph from Slack, Jira, and calendar.
💡 Run /usermodel to build a profile of your priorities and working style.
⚠️ Embedding server stopped⏳ Installing embedding server📦 Embedding server not installed🔄 Starting embedding server — …⏸ Embedding server stopped
Last error: . KG building and semantic search need a running local embedding model. Restart or switch to OpenRouter in Settings → AI.TARS uses a local model for embeddings (~10 GB one-time download). Install now, or switch to OpenRouter in Settings → AI.Downloading dependencies. This runs once and can take a few minutes.Start the server to enable embeddings, or switch to OpenRouter in Settings → AI.
⚠️ Claude CLI authentication failed
Open a terminal and run claude login, then send any message to retry.
Until you do, calls fall back to OpenRouter (billed to your API key).
Run the embedding model on your AMD GPU (~2–3× faster, frees the CPU).
One-time: grants this user GPU compute access (you'll get a system password prompt).
Access granted — log out and back in to finish.
Working…
Worker LLM
Background tasks: KG triple extraction, compact summarisation, session signals. Separate from the main chat model.
Ollama:● Running● Starting…Installed, not runningNot installedInstalling…
Models
Pick model / judge / thinking per turn using the three spinners next to the chat box
(click or scroll each). Judge runs a fresh-context verifier after each response — set it
to a tier to enable, or Off to skip.
Behaviour
Atlassian (Jira / Confluence)
Connect your Atlassian account to access Jira and Confluence.
Connected to
Meeting Categories
Customize categories for organizing meeting notes via the /recap command.
Slack
Connect your Slack workspace to search messages and send DMs as yourself.
Connected to
Configure which Slack scopes to search.
Miro
Connect Miro to let the AI read your boards for research and insights.
Connected to Miro
Bitbucket
Connect Bitbucket to let the AI search repos, PRs, branches, and pipelines.
Connected to Bitbucket
GitLab
Connect GitLab to let the AI search projects, MRs, branches, and pipelines.
Connected to GitLab
Todo Source
Choose where to fetch your tasks from.
Jira Todos
Connect your Atlassian account above to configure Jira todos.
Slack Lists Configuration
Enter the Slack List ID to use as your todo source.
Connect to Slack above to use Slack Lists as your todo source.
Email
Team / Org Structure
Add your manager and team members so the AI can answer questions about your team, reporting lines, and org structure.
Loaded from Drive:
Calendar Sidebar
Font
Notifications
World Clock
Desktop only. Displays up to 4 timezone clocks in a trading-floor style bar below the header.
Voice Controls
Voice Output
Speech Recognition
Wake Word
Tool Permissions
Tools permanently approved to execute without confirmation. Click × to revoke.
Proactive Interventions
Automatically surface priorities, stale follow-ups, and actionable items at session start.
Claude CLI Provider
Route the main agent through your locally-installed claude CLI instead of OpenRouter, so calls bill to your Claude subscription.
Embeddings, the worker model, and the Slack rewriter stay on OpenRouter regardless. Single-user, on-machine usage only — see
docs/claude-cli-electron-provider.md for the full TOU rationale.
✅ Logged in:
⚠ Logged in via Anthropic Console (API-billed, not subscription) —
Not signed in. Click Sign in below or run claude auth login in a terminal.
❌ claude CLI not found on PATH. Install Claude Code first, then come back.
Checking…
On a chat turn: TARS spawns claude -p with strict MCP scope, streams partial messages back, and
routes all TARS tool calls back through the renderer's tool-router (so OAuth + confirmation modals still work).
Falls back to OpenRouter automatically if the CLI fails mid-turn.
Prolog Bridge
Skill Editor / Drive / Backups
Edit the assistant's behavior, prompts, patterns, and tools. Changes persist to Drive.
View and restore previous versions of your data files from Google Drive.
Data Retention
Cue Light
Requires desktop app — Cue Light runs a local Gemma model via Ollama.
Recommended models
Installed
Cancelled. Restart the download to resume from where it stopped.
MCP Servers
Connecting...
Connection failed
Grafana Loki
Query application logs via Loki. The AI can search logs, check errors, and investigate incidents.
Loki configured
Slack Alerts
Connect Slack above to use alerts.
Monitor channels and get alerted when messages match your alert definitions.
#
Select up to 20 channels. Currently selected.
Alert sensitivity is now configured per-alert in the Alerts tab.
build 743ddbc
Comments