One agent. Three capabilities. Zero bloat.
LightEDR is a single lightweight agent feeding a multi-tenant control plane. Here's what it does - and what it deliberately doesn't.
Real-time behavioural detection.
A curated ruleset mapped to MITRE ATT&CK, updated as new techniques emerge. Extend with your own Sigma rules - no proprietary DSL.
- Curated, not kitchen-sink.
~400 rules, not 40,000. Tuned to fire when it matters.
- Sigma-native.
Import Sigma rules as-is. Export them too - nothing stays locked in.
- Opt-in ML scoring.
For teams who want it. Off by default - we don't believe in selling confidence scores you can't explain.
Contain in a single click.
Every response action is atomic, audited, and reversible. Run from the console, the API, or a playbook - actions land on the endpoint in under 500ms.
Isolate host
Network quarantine with a single exception for your console. Reversible.
Kill process
Tree-aware - kill the parent, kill the descendants. Or just one PID.
Quarantine file
Move to isolated store, checksum, preserve for forensics.
Custom playbooks
Stitch actions together in YAML. Version-controlled, reviewed, auditable.
The events you need, nothing you don't.
Process, network, file, and authentication events - deduplicated at the agent, enriched at ingest, queryable in standard SQL for 90 days hot and one year warm.
The only EDR with Zero Trust built in - not bolted on.
Every competing EDR authenticates an agent once and trusts it forever. LightEDR follows NIST SP 800-207: a Policy Decision Point continuously scores every active connection; Policy Enforcement Points act on that score in real time. Authentication is a live signal, not a historical fact.
CSR enrolment - private key never transits
On first run the agent generates a keypair locally using mbedTLS. It sends only a CSR to the triage server, authenticated with a single-use 24h enrolment token. The triage engine signs the CSR and returns only the certificate. The private key is generated on-device, stored at /etc/light_edr/agent.key (chmod 600), and never transmitted over any network.
Mutual TLS + replay-resistant auth
Every connection after enrolment is mTLS. The agent presents its certificate; the triage engine verifies it against the CA. Every auth frame carries a timestamp and a random UUID nonce validated against Redis - duplicate nonces within the window are rejected outright.
Cryptographic challenge-response every 5 minutes
The triage engine sends a random nonce to every active session. The agent must sign SHA-256(nonce ‖ agent_id) with its private key. This proves the key is still resident on the enrolled endpoint throughout the session - not just at connect time.
Instant revocation via Redis pub/sub
When a cert is revoked, a Redis pub/sub message closes every active WebSocket for that agent immediately. Revocation does not wait for the next reconnect. Fail-closed: if the revocation DB is unavailable, connections are denied - not allowed.
Tamper-evident hash-chained audit log
Every auth event writes a record containing SHA-256(prev_id:prev_ts:prev_type). A periodic chain-walk job detects any gap, reorder, or modification. The database user has INSERT + SELECT only - no UPDATE or DELETE on the audit table.
The platform your AI agents operate on.
Most security platforms are adding AI as an afterthought - a chatbot layered over a legacy UI. LightEDR was designed from the ground up to be operated by AI agents. The MCP server, webhook system, and scoped service account model are not integrations. They are first-class features.
Full AI & Agents feature page →Model Context Protocol - 13 tools, zero prompt engineering
The MCP server (pip install ledr-mcp) exposes every major platform operation as a named, described tool. Any MCP-compatible client - Claude, Cursor, Zed - connects with two environment variables and a JSON config block. No custom SDK. No proprietary protocol.
list_alertsget_alertget_alert_contextacknowledge_alertinvestigate_alertisolate_agentunisolate_agenthunt_eventsget_mitre_coverageBring your own agent - model-agnostic by design
You are not buying a "Claude integration" or a "GPT feature." You are buying a security platform that any agent can operate. When better models ship, you upgrade the model. The security stack stays the same.
Scoped service account keys mean the agent has exactly the permissions it needs - analyst read access for investigations, agents:manage only if host isolation is required. Principle of least privilege, enforced at the API layer.
Reactive, not polling
Webhook channels wake your agent when a rule fires. POST to your endpoint, your agent investigates and responds, then goes back to sleep. Exponential backoff retry, HMAC request verification, async-safe. No polling. No wasted tokens.
Zero Trust under the hood
The AI agent connects over HTTPS with a scoped service account key. The endpoints it calls are the same ones your analysts use - fully authenticated, rate-limited, and written to the audit log. Every action the agent takes is visible to humans. There is no back-channel.
Try LightEDR on 50 endpoints, free.
Sandbox tenant, full feature set, 14 days. No credit card.