Summary
Summary
This session was a more casual, chat-driven talk from Lyzr’s Product Lead on how the company used its own agent framework, GitAgent, to support its Series B fundraise. The centerpiece is SivaClaw, an investor-relations agent built on GitAgent (Lyzr’s version-controlled, secure runtime inspired by OpenClaude’s architecture but with stronger governance and security). Rather than a deterministic workflow, SivaClaw was built with knowledge files, a soul.md, and a rules.md that let it answer dynamically and stay within guardrails, backed by Lyzr’s Responsible AI policies to catch prompt injection or malicious probing.
SivaClaw operated in two main phases. In pre-due diligence, a public-facing version of the agent let anyone (investors included) ask about Lyzr’s product, team, traction, and history, generating personalized investment memos on request. This filtered roughly 200+ investor conversations down to about 40-50 real discussions and around 30-35 substantive product due-diligence conversations before the founder ever got on a call, saving significant time. In the due-diligence phase, the agent was embedded directly in the data room, answering detailed questions on financials, ARR, customer lists, cap table, and GTM, always citing the source knowledge file so investors could verify claims and the agent wouldn’t hallucinate on numbers. The founder noted the same approach was used successfully in their $8M Series A and is being repeated for Series B, with plans to eventually open-source the data room product (SivaClaw itself is already open source, built on GitAgent).
Key Takeaways
- Agents can compress the fundraising funnel. By handling repetitive, informational investor questions upfront, SivaClaw cut roughly 200 investor touchpoints down to 30-35 real due-diligence conversations, letting the founder walk into calls already informed rather than starting from scratch each time.
- Structured data plus citations prevents hallucination risk. For financial and factual claims (ARR, gross margin, customer counts), the agent was designed to always cite the source knowledge file, keeping it auditable and grounded rather than creatively “pitching” like a human would.
- The harness matters more than the model. The founder credited GitAgent’s architecture (knowledge files, rules.md guardrails, Responsible AI policies) as what made a public-facing, high-risk agent (open to malicious probing) usable at all, reinforcing the harness engineering theme from the earlier webinar.
Summary
This session was a more casual, chat-driven talk from Lyzr’s Product Lead on how the company used its own agent framework, GitAgent, to support its Series B fundraise. The centerpiece is SivaClaw, an investor-relations agent built on GitAgent (Lyzr’s version-controlled, secure runtime inspired by OpenClaude’s architecture but with stronger governance and security). Rather than a deterministic workflow, SivaClaw was built with knowledge files, a soul.md, and a rules.md that let it answer dynamically and stay within guardrails, backed by Lyzr’s Responsible AI policies to catch prompt injection or malicious probing.
SivaClaw operated in two main phases. In pre-due diligence, a public-facing version of the agent let anyone (investors included) ask about Lyzr’s product, team, traction, and history, generating personalized investment memos on request. This filtered roughly 200+ investor conversations down to about 40-50 real discussions and around 30-35 substantive product due-diligence conversations before the founder ever got on a call, saving significant time. In the due-diligence phase, the agent was embedded directly in the data room, answering detailed questions on financials, ARR, customer lists, cap table, and GTM, always citing the source knowledge file so investors could verify claims and the agent wouldn’t hallucinate on numbers. The founder noted the same approach was used successfully in their $8M Series A and is being repeated for Series B, with plans to eventually open-source the data room product (SivaClaw itself is already open source, built on GitAgent).
Key Takeaways
- Agents can compress the fundraising funnel. By handling repetitive, informational investor questions upfront, SivaClaw cut roughly 200 investor touchpoints down to 30-35 real due-diligence conversations, letting the founder walk into calls already informed rather than starting from scratch each time.
- Structured data plus citations prevents hallucination risk. For financial and factual claims (ARR, gross margin, customer counts), the agent was designed to always cite the source knowledge file, keeping it auditable and grounded rather than creatively “pitching” like a human would.
- The harness matters more than the model. The founder credited GitAgent’s architecture (knowledge files, rules.md guardrails, Responsible AI policies) as what made a public-facing, high-risk agent (open to malicious probing) usable at all, reinforcing the harness engineering theme from the earlier webinar.