Organizational memory
Bidderops’s AI agents learn about your organization over time through organizational memory — a semantic memory layer that captures lessons from past bids, project experience, intelligence reports, and your direct corrections. When an agent runs, it automatically retrieves the most relevant memories and uses them as context, producing advice that is more specific to your organization’s strengths, weaknesses, and history.
How it works
Every time your team records a bid outcome, completes an intelligence report, adds a business reference or CV, or corrects an agent output, the system stores a compact summary of that event as an organizational memory. These summaries are embedded using a lightweight AI model that the platform manages — no configuration needed on your part.
When you later ask an agent for advice (Go/No-Go, Capability Match, Pipeline Report, or Bid Intelligence), the system searches your organization’s memories for the most relevant context and includes it in the agent’s prompt. For example, if you are evaluating a water-supply opportunity in Kenya, the agent might recall that you won a similar project there last year and that your ISO 9001 certification was a key differentiator.
What gets remembered
| Source | When stored | Example |
|---|---|---|
| Bid outcomes | When you record a win or loss | ”Lost transport bid in Nigeria — our price was 15% above the winner” |
| Intelligence reports | When the intelligence agent completes a dossier | Key people, competitors, and project summary |
| Business experience | When a reference is added or imported | Project title, sectors, countries, contract value |
| Personnel | When a CV is added or imported | Name, expertise areas, sector and country experience |
| Notes | When you add a note to an opportunity | The note content with opportunity context |
| Corrections | When you correct an agent output | Your specific feedback for future runs |
Correcting agent outputs
Every agent output panel (Go/No-Go, Capability Match, Intelligence, Pipeline Report) includes a Correct this result link. Clicking it opens a text field where you can tell the AI what it got wrong or what it should remember for next time.
Corrections are the highest-value memory type because they capture your explicit preferences and domain knowledge. For example:
- “Always highlight our ISO 9001 certification in capability assessments”
- “We have in-country presence in Kenya through our Nairobi office”
- “For UNDP-funded projects, emphasize our HACT compliance”
These corrections are stored permanently and surface whenever a future agent run touches the same sector, country, or topic.
Which features use memory
| Feature | Memory types used |
|---|---|
| Go/No-Go Advisor | Bid outcomes, corrections, agent interactions |
| Capability Match | Bid outcomes, experience, personnel, corrections |
| Pipeline Report | Bid outcomes, corrections |
| Bid Intelligence | Intelligence reports, bid outcomes |
Privacy and isolation
Organizational memory is strictly scoped to your organization. Your memories are never visible to other organizations, and the similarity search is always filtered by your organization ID. The embedding model runs on the platform’s infrastructure — it does not use your AI provider key.
Availability
Organizational memory is available on plan tiers that include the semantic memory feature. Contact your administrator to check your plan’s included features.