Setting Up Agent Insights
Connect an agent
From Agent Insights → Get Started, follow the prompt to connect the first coding agent your team uses (Claude Code, Cursor, Codex, or others).
Invite your team
Add the teammates whose agent usage you want visibility into. Each person needs to complete the CLI step below to start showing up in the dashboard.
Each developer connects the CLI
pip install entelligence-cli
entelligence auth login
# One-time: configure the LLM used for local analysis
entelligence insights setupinsights setup walks you through: which LLM provider to use for local analysis (any provider your CLI can reach — Anthropic, OpenAI, Google, or a local Ollama model via --base-url), which model, and optionally a separate (often cheaper) model just for digest generation. It finishes by downloading a small local embedding model for clustering, offering a test of the connection, and offering to install a scheduled sync (see below).
# Index local session files (free, no LLM calls)
entelligence insights index --source local --hours 8760
# Push computed metrics up to your org's dashboard
entelligence insights syncinsights sync only pushes an allow-listed set of pre-computed metrics — token counts, tool-call names and counts, efficiency scores — never raw session transcripts, and never tool input/output (where code, commands, or secrets could live).
Prefer not to run analysis through an LLM at all? insights index alone is enough to start syncing coarse activity metrics — insights analyze (which does use an LLM, and shows you a cost estimate before running) adds the qualitative friction/quality layer on top.
Keeping data fresh automatically
Rather than remembering to run sync yourself:
entelligence insights schedule --at 09:00Installs a daily job (launchd on macOS, cron on Linux) that runs entelligence insights sync every morning at the time you specify. Remove it with entelligence insights schedule --remove.
This is a different schedule than cinsights' own local index+analyze scheduler — entelligence insights schedule only supports --at / --remove and always drives sync, even though the underlying cinsights tool has a more feature-rich schedule install/status/uninstall of its own. See the CLI reference for the distinction.
Optional: broader data sources
- Entire.io — git-based checkpoints for cross-agent, cross-machine coverage across a team.
- Arize Phoenix — centralized trace observability if your org already runs Phoenix.
Set either as the --source on index/analyze instead of local if your org uses them.
Running it yourself locally
Every developer can also use the underlying engine directly, independent of the team dashboard:
# Generate a project or personal digest
entelligence insights digest project <my-project> --days 30
entelligence insights digest user <user-id> --days 30
# Launch the local web dashboard
entelligence insights serve # http://localhost:8100See entelligence insights --help for the full command list, or the CLI reference.
Reminding unsynced teammates
Admins can send a one-click reminder from the Agent Insights dashboard to anyone invited but not yet syncing data.
Troubleshooting
| Symptom | Likely cause | Fix |
|---|---|---|
| Teammate doesn't show up in the dashboard | They haven't run insights sync yet | Send a reminder from the dashboard, or have them run the 3-command setup above |
insights sync reports nothing to push | No new sessions since the last sync | Run insights index again first, or use --all to force a full re-push |
insights analyze cost estimate looks high | Large --limit/uncapped session window | Lower --limit, raise --min-score to only analyze higher-signal sessions, or use --tasks-only to skip the most expensive step |
| Want to see exactly what would be sent before syncing | — | Run entelligence insights sync --dry-run first |