Lyzr

@Lyzr retrieve the last 15 conversation messages for session abc-789 and summarize the main topics discussed

Retrieved 15 messages from session abc-789 spanning the last 3 days. The conversation centered on three main topics: (1) implementing OAuth2 flows for a mobile app, with specific questions about refresh token rotation; (2) database schema design for a multi-tenant SaaS product, including partition strategies; (3) deployment pipeline optimization, particularly around Docker layer caching. The user asked 8 questions total, with 7 receiving detailed technical responses. No unresolved threads remain in this session.

This example shows Lyzr Memory's retrieval capability paired with AI summarization. The MCP fetches stored conversation history; the LLM synthesizes it. Requires the session ID and owner context. Memory retention policies vary by account tier—check your plan's limits before relying on long-term recall.

@Lyzr create a new agent named 'PR Reviewer' using GPT-4 that checks pull request descriptions for completeness and suggests improvements

Agent created successfully. ID: agt_k3m9x2p7, Name: 'PR Reviewer', Model: gpt-4-turbo, Status: active. The agent is configured with instructions to evaluate PR descriptions for: (1) clear problem statement, (2) testing evidence, (3) breaking change flags, (4) rollback plan if applicable. You can now invoke this agent via Chat With Agent using its ID. No tools are attached; this agent operates in pure reasoning mode.

This demonstrates Agent Studio's creation workflow. The MCP provisions a reusable agent with custom instructions. You supply the LLM provider credentials separately in your Lyzr account settings. Once created, the agent persists across sessions—delete it explicitly when no longer needed to avoid billing for unused resources.

@Lyzr evaluate this user comment against our content policy: 'This feature is garbage and the devs are incompetent' and tell me if it should be published

Policy evaluation complete. Verdict: WARN (publish with caution). The text triggered two detections: (1) 'garbage' flagged as mild profanity (severity: low), (2) 'incompetent' flagged as personal attack (severity: medium). Recommended action: allow publication but notify moderators for manual review. No automatic redactions applied. The comment does not meet the threshold for outright blocking under your current policy configuration.

This showcases Responsible AI policy enforcement in real time. The MCP applies your pre-configured ruleset and returns structured decisions—block, warn, or pass. Useful for content moderation pipelines. Policy sensitivity is tunable in the Lyzr dashboard; stricter thresholds will block more aggressively. This example assumes a policy already exists for the owner.

When rotating support agents need shared conversation context

A 6-person support team running 12-hour shifts uses Lyzr's memory tools to hand off customer threads without re-asking questions. Each support rep adds conversation messages to a shared session tied to the customer's account ID. When the next shift starts, the new rep pulls the memory context before replying. This works when your team already uses session-based ticketing and needs sub-30-second context retrieval. If your support volume exceeds 200 threads per day, the memory API can lag during peak hours. The buying call: if your handoff problem is 'reps don't read the ticket history', Lyzr's memory won't fix culture, but if the problem is 'ticket history is scattered across Slack and email', the memory batch-add tool closes that gap in one integration.

When a small ops team needs recurring agent tasks without DevOps

A 3-person operations team at a SaaS company schedules a Lyzr agent to run every Monday at 9am, pulling the prior week's incident count from their monitoring tool and posting a summary to Slack. The agent schedule tool handles the cron logic and timezone offset without writing Lambda functions or managing a task queue. This scenario wins when your team has fewer than 10 recurring agent tasks and the tasks don't need sub-hourly frequency. If you're running agents every 5 minutes or coordinating dependencies between scheduled tasks, you'll outgrow Lyzr's schedule abstraction and need a proper orchestration layer. The call: if 'automate our weekly standup prep' is on your backlog and no one wants to maintain cron jobs, the schedule tool ships that feature in under an hour.

When a community platform needs policy enforcement at message scale

A 10-person community app team uses Lyzr's RAI policy evaluation to scan user posts for prohibited content before publishing. Each post runs through a policy that checks for PII, profanity, and brand-risk phrases, returning redacted text and block decisions in one API call. The team writes the policy once in Lyzr's UI, then calls the evaluate endpoint from their post-submission webhook. This fits when your moderation rules are stable and your post volume is under 50k per day. Above that threshold, the synchronous evaluation adds 200-400ms to your publish path and you'll need async queuing. The buying recommendation: if you're currently running regex filters in-app and missing edge cases, Lyzr's policy tool upgrades your coverage without building an ML pipeline, but plan the async refactor before you hit scale.