Bid intelligence

Bid intelligence produces a research dossier on a single opportunity to help you decide how to position and team for the bid. It lives on the opportunity’s Competitive Edge tab, alongside capability match. The dossier and its deep dives are written by the Researcher, one of your agents.

What the dossier covers

The agent researches four areas:

1. Project deep-dive — the background, objectives, scope, funding context, and the strategic context behind the specific procurement.
2. Key people — named individuals likely involved (project directors, programme managers, evaluators, contracting-authority contacts) and why each matters.
3. Competitors and partners — firms likely to bid against you, and firms that would make strong joint-venture or subcontract partners.
4. Suppliers and consultants — specialist firms and individual consultants who could strengthen your bid.

The report comes back as a structured dossier with a project summary, key facts, people, competitors/partners, suppliers, and a list of sources. When the model runs through OpenRouter, the pages OpenRouter’s web search cites are added to the sources too.

Using the RFx analysis

If you have already run the RFx analyzer on the bid’s tender documents, the dossier and every deep dive use the saved analysis. The agent receives the scope of work and the evaluation criteria, and uses them to focus on who is likely to evaluate the bid, which competitors fit those criteria, and which partners would close your gaps. Without a saved analysis the agent works from the opportunity’s details alone, so analyze the documents first for the best results. The Competitive Edge tab tells you which case applies.

Deep dives

From any section of the dossier you can request a deep dive — a focused, in-depth analysis of just that aspect (the project, the people, the competitors, the partners, or the suppliers), going far beyond the summary. Deep dives run on the model selected at the top of the panel, the same one that writes the report.

Grounded vs. ungrounded

Bid intelligence is most valuable when it is grounded in live web search, so it can cite current, verifiable facts. Whether grounding happens depends on the AI provider you have chosen for this feature:

• Anthropic, OpenAI, Google, OpenRouter support web-search grounding.
• Local models cannot perform web search, so the report falls back to ungrounded generation. It is still produced, but it draws only on the model’s own knowledge, and the report is flagged grounded: false.

See the AI providers page for the full grounding support matrix, and the Local LLM page for what this means when running on your own hardware.

Where it runs

Generation runs as a durable background job, so you can navigate away and come back — the report keeps working and is marked complete when finished. Each finished report is also added to your organizational memory, so later agents can draw on it.

When a report fails

If a run fails, your previous report stays in place and the panel shows why. The most common reasons:

• The reply was cut off at the model’s output limit. A full dossier is long; the agent allows up to 32,768 tokens on OpenRouter and Anthropic (less on OpenAI, Google and local models, whose models cap output lower). Try a model with a larger output limit.
• The model used its whole output limit without writing a reply. Reasoning models can spend the entire limit thinking. On OpenRouter their thinking is capped at 8,192 tokens; if it still happens, pick a model that reasons less.
• No JSON found in response. The model wrote prose instead of the structured dossier. Smaller models do this more often; a larger model usually fixes it.