Governed AI agents for IT operations
SWARM works through the systems you already run. It writes a plan before it acts, stops for approval where your policy says so, and records what worked. Self-hosted, on your data.
One assistant across
your whole estate
SWARM connects to 83 types of system and acts through them with the permissions you grant, per integration.
Infrastructure Automation
SSH, Ansible, Terraform, CloudFormation — describe the change in plain language; SWARM plans it and runs it through your existing automation.
Observability & AIOps
Splunk, Elasticsearch, Kibana and more — search runs at the source, anomalies become incidents, and incidents are correlated across tools.
Guardrails & Approvals
Policies are checked before every tool call, high-risk actions wait for a named approver, and every step is written to the audit log.
Document Intelligence
Your runbooks, wikis and documents answer questions with citations; hybrid search and a grounding check keep answers tied to the source.
Code & DevOps
GitLab, GitHub, CI/CD pipelines — review pull requests, manage branches, trigger pipelines and record deployments.
Multi-Tenant IAM
One installation serves isolated tenants — role-based access, read/write/admin permission per integration, SSO (SAML, LDAP, OAuth) and a full audit trail.
A written plan,
checked at every step
A plan-first orchestrator decides what to do before anything runs. Guardrails and approvals sit between the plan and your systems, and every outcome is recorded.
What the agents do
Each agent has one job. All of them run under the same permissions, guardrails and approvals as a person using SWARM.
Planner
Turns a request into a written plan — steps, tools, checks — before anything runs, and can hand self-contained sub-tasks to sub-agents on the model set for that area.
Detection agent
Polls observability sources such as Splunk and Elastic for anomalies and opens an incident.
Triage agent
Collects evidence, sets severity, and opens the ticket in your ITSM with the evidence attached.
Remediation agent
Picks a runbook, simulates it, checks the guardrails, waits for approval where required — then runs it or escalates.
SRE agent
Adds the incident timeline and the state of the affected service-level objectives.
Introspection agent
Watches SWARM itself — agent, connector and queue health — so a stalled pipeline is noticed.
From request to result
in four steps
-
Plan
The request becomes a written plan: the steps, the tool each one uses, and how the result will be checked. You see the plan before anything runs.
-
Approve
Guardrail policies are evaluated before every tool call. A high-risk action — or any action your policy names — is staged for a named approver instead of running.
-
Act
Tools run through the integrations you connected, each with its own read, write and admin permission. Every step is checked, and a failed step triggers a correction.
-
Learn
Outcomes are recorded, answers your team rates well are remembered for similar requests, and a plan that worked is reused for the same request.
Natural language.
Real infrastructure.
Describe the outcome. SWARM plans it, asks where it must, and does it. (Illustrative session.)
SWARM
Chat
Install a web application on this host, configure it to listen on port 443, use Let's Encrypt to generate a certificate, create a task that renews the certificate when it has 7 days left, check the certificate's expiry, and create a ticket with the details of this session.
Plan · 5 steps across SSH, the scheduler, certificates and ServiceNow
- Install the web application
- Configure HTTPS on port 443 with Let's Encrypt
- Schedule certificate renewal
- Check the certificate's expiry
- Open a ServiceNow ticket with the details
shell:run_scriptInstalling the web application and its dependencies…Application installed, service enabledshell:run_scriptConfiguring HTTPS on port 443 and requesting a Let's Encrypt certificate…Certificate issued, HTTPS listener activesystem:schedule_taskScheduling renewal when the certificate has 7 days left…Renewal task created — runs daily at 03:00system:cert_manageChecking the certificate's expiry…Certificate valid for 89 daysservicenow:createCreating a ticket with this session's transcript and change log…Incident created in ServiceNow
All 5 steps completed. The application is live on :443 with automated certificate renewal.
Your agents are portable.
By design.
SWARM conforms to SC-006 (AAIF) v3.4.0 — the Autonomous Agent Interchange Format, an open Schema Commons standard for vendor-neutral agent definitions.
Personas and playbooks can be exported as schema-valid AAIF documents and imported from any conforming runtime — with capability negotiation, dry-run previews, and lossless round-trips. No lock-in, verifiably.
- Import & export via the AAIF REST API (personas ↔ agent definitions, playbooks ↔ orchestrators)
- Capability negotiation against the published SC-006 capability registry
- Public self-certified conformance endpoint — Enterprise level, producer & consumer
curl https://swarm-software.ai/.well-known/aaif-conformance.json
Ready to deploy
your SWARM?
Self-hosted. No vendor lock-in. Full control over your data and agents.
Or get in touch
Questions, demos, or partnership inquiries — we'd love to hear from you.