Agent with connected tools

Ask the interactive shell in plain language — it can pull live data from your integrations before answering.

In the interactive shell (opensre), ask a free-form question. If the answer needs data from a connected integration (GitHub, Datadog, Sentry, Grafana, and others), the agent calls those tools and answers from their results.

You do not pick a tool or run a slash command for this. Just ask.

Example

> can you check the github issues for the windows crashes in https://github.com/Tracer-Cloud/opensre?
· gathering via GitHub · search issues — windows crashes…
There are 3 open issues mentioning Windows crashes. The most recent, #482
"App crashes on launch (Windows 11)", reports a startup segfault after the
2.4.0 update…

> what do the datadog logs say about payment-service errors in the last hour?
· gathering via Datadog · query logs — payment-service errors…
Found 847 error log lines in the last hour for service:payment-service.
Top pattern: ConnectionTimeout to postgres-primary…

The dim progress line shows which integration and tool ran.

For GitHub, you can put the repo URL in the question, mention it earlier in the session, set GITHUB_REPOSITORY, or let OpenSRE use the current checkout’s origin remote when GitHub is connected.

How it works

1. You ask a free-form question (not a /slash command).
2. OpenSRE decides whether live data would help and calls the matching chat tools.
3. The ReAct loop is bounded so the shell stays responsive. Tool results are returned to the same model before it answers.
4. Tools are read-only data fetches — they run without a per-call confirmation (same idea as an investigation lookup, not a mutating setup wizard).
5. Progress shows on the dim · gathering via <source> · <tool> — <hint>… line while tools run.
6. The agent answers using that evidence and recent chat context.

What happens when something goes wrong:

Situation	Result
Ctrl+C during tool use	The turn stops
Tool or integration unavailable	The agent reports what is unavailable
No usable tools for the question	The agent answers without a tool call
Full RCA needed	Use /investigate — this path is for Q&A with live lookups, not a full multi-stage investigation

This is not the same as the investigation pipeline. Investigations run a separate multi-stage RCA and deliver a report.

Setup

You need an LLM and at least one integration for live tool calls.

1. LLM

opensre onboard
# or set LLM_PROVIDER plus the provider API key — see LLM providers

In the shell: /model show, /auth login <provider> if needed.

2. Connect integrations

opensre integrations setup datadog
opensre integrations setup github
# or in the shell:
/integrations setup
/integrations list

Only connected and verified integrations contribute tools.

3. Check that tools are available

/health          # which integrations look healthy
/verify datadog  # one service
/tools           # registered tools

If /health is red or /tools is empty for that source, fix setup first.

4. Ask

what do the datadog logs say about payment-service errors in the last hour?

You should see a tool progress line when a tool runs. If you do not, the question may not need live data or that integration may be unavailable.

With no integrations connected, behavior is unchanged: answers come from text and docs only. Connect a source and matching questions start returning live data.

You ask	Typical source
Check GitHub issues for Windows crashes	GitHub
What do Datadog logs say about payment-service?	Datadog
Recent Sentry errors on checkout?	Sentry
Error rate in Grafana for prod-api?	Grafana

Related

• Shell commands
• Integrations overview
• Investigations overview
• GitHub