Lyzr

What it is

Lyzr is an enterprise platform for building, testing, deploying, and governing AI agents. You build agents in three ways: visually in Agent Studio, in code with the Python/TypeScript Agent Development Kit, or by describing the agent in plain language to a builder called Architect. Agents can work alone, under a Manager Agent that coordinates specialist sub-agents, or inside a Superflow, Lyzr’s deterministic multi-step pipeline. The platform also bundles a RAG layer with knowledge graphs and text-to-SQL, a memory module called Cognis, voice agents over Twilio, Telnyx, and Plivo, and a governance set: PII redaction, prompt-injection blocking, a Hallucination Manager, Agent Eval test generation, step-level tracing, and SSO with audit logs. It is model-agnostic. Each agent can use OpenAI, Anthropic, Google, AWS Bedrock, Groq, or Perplexity, or any OpenAI-compatible endpoint you host. It works as an MCP client, and it exposes Cognis as an MCP server.

The closest category leader is Microsoft Copilot Studio. The difference is deployment. Lyzr runs as SaaS, inside your own VPC, or as a hybrid with cloud orchestration and on-premises data connectors, and it does not assume a Microsoft tenant.

Lyzr was founded in 2023 and is headquartered in the New York area. Accenture Ventures invested in October 2025 to push the platform into banking and insurance. In March 2026 Accenture led a $14.5M Series A+ at a $250M valuation, according to Bloomberg. In July 2026 Bloomberg reported that Lyzr was raising about $100M at roughly a $500M valuation and had used its own agent to handle investor outreach. Lyzr has not named a lead investor, and coverage disagrees on whether the round has closed. Treat the $500M figure as company-stated.

Why it shows up in ops stacks

Lyzr shows up in regulated-industry shortlists for two reasons. First, the governance layer ships with the platform: guardrails, hallucination checks, evals, and audit logs are not separate products. Second, the VPC option keeps data and compute inside your own cloud, at roughly half the SaaS per-run price. Accenture’s investment was explicitly aimed at banking and insurance workflows such as claims processing, loan approvals, and GRC audits. For a RevOps, legal ops, or recruiting team inside a bank or insurer, that combination is often what gets an agent past security review.

Pricing reality

Lyzr publishes its rates, which few enterprise agent platforms do. Billing is in credits, with 1 credit equal to $1. A run is one request from input to output, however many agents or tool calls it uses inside. Per-run rates on Lyzr’s SaaS cloud, with the customer-VPC rate in brackets:

• Single agent — $0.06 per run ($0.03)
• Manager Agent, 1–4 sub-agents executed — $0.18 ($0.09); 5–8 is $0.36, 9 or more is $0.54
• Superflow, up to 10 nodes — $0.30 ($0.18); steps up per 10 nodes, plus $0.02 per node beyond 30
• Voice — $0.09 per minute ($0.06)

LLM tokens are passed through at cost, or you bring your own model and pay the provider directly. Subscription tiers cap the build surface. Community is free with 5 credits a month and 10 agents. Starter is $19 a month with 20 credits and 15 agents. Pro is $99 a month, or $79 billed annually, with 100 credits, 25 agents, and 1 GB of RAG storage. Every self-serve tier has one builder licence and 7-day logs. On-prem deployment, Agent Eval, the Hallucination Manager, Responsible AI controls, and full orchestration are Enterprise only, at custom pricing.

What teams actually pay is set by run volume, not by the plan. A manager-agent research flow at 10,000 runs a month costs $1,800 in platform fees on SaaS, or $900 in your VPC, plus model tokens. Lyzr’s own worked example of that VPC case adds about $400 of LLM cost, for $1,300 a month. Pro’s 100 credits cover about 1,650 single-agent runs before tokens, so any production workload means top-ups, sold one-time from $10 with no expiry, or an Enterprise contract. Our estimate: budget Enterprise as at least a five-figure annual commitment. Lyzr publishes no Enterprise price. You need that tier for the governance features that are the reason to pick Lyzr in the first place.

Best for

Platform or automation leads at banks, insurers, and other regulated mid-to-large enterprises who need several governed agents in production inside their own cloud, with model choice per agent and per-run pricing they can forecast.

Not for: a team that wants one agent for one workflow this quarter, a non-technical team without a builder, or a Microsoft 365 shop whose users already hold Copilot licences. That last group gets employee-facing agents with no extra credit spend in Copilot Studio.

Versus the alternatives

• Microsoft Copilot Studio — the market-share leader for building agents in-house, at $200 a month for 25,000 credits. Pick it when your data and identity live in Microsoft 365. Pick Lyzr when you need VPC deployment outside Azure or a model Microsoft does not host.
• Salesforce Agentforce — the other share leader, at $0.10 per standard action on Flex Credits. Pick it when the agent works mostly on CRM records. Pick Lyzr when the workflow spans systems Salesforce does not own.
• n8n — the fastest-growing entrant, valued at $5.2B in May 2026. Pick it when ops engineers want self-hosted, execution-priced flows and will build their own guardrails.
• Relevance AI, Dust, Sim — cheaper self-serve builders for teams that do not need on-prem. WRITER fits non-engineers who mostly need governed content agents. Kore.ai fits contact-centre automation at scale.

Watch-outs

• Manager and Superflow runs step up silently. A manager that starts calling a fifth sub-agent doubles its per-run price, and loops and retries count as Superflow nodes. Guard: set the budget caps in Lyzr’s cost controls before go-live, and review the per-agent cost split in the billing view weekly for the first month.
• Self-serve tiers lack the governance you are buying. Evals, the Hallucination Manager, and on-prem sit behind Enterprise. A Pro pilot does not test them. Guard: run the pilot on an Enterprise trial or sandbox agreement, and put the security review on that configuration, not on Pro.
• MCP servers must be publicly reachable. Lyzr’s docs require a public URL for each server, which exposes internal systems. Guard: put internal MCP servers behind an authenticated gateway with OAuth and per-user audit logging, or use the hybrid deployment so connectors stay on-premises.
• Most proof is vendor-reported. Lyzr’s G2 listing carries about 48 reviews, and its revenue figures differ by source. Guard: ask for a reference customer in your industry and deployment model, and write workflow-definition export into the contract.