Skip to main content

​
How the agent works

Orbid is a session-based agent. You provide files and instructions; Orbid chooses tools (scan, match, fill, generate, research, …), returns intermediate results, and often stops for your judgment before high-stakes steps.

​
The loop (user view)

This is the same pattern as other native agents: delegate → tools → review → continue.

​
Sessions

  • One session ≈ one bid effort
  • Holds conversation, attached files, match state, and drafts
  • Switching product models may use branches / re-match—do not blend two models’ evidence silently
See Agents and sessions.

​
Context the agent can use

  • Session uploads — Primary tender truth
  • Library — Product and supporting files
  • Knowledge — Org lessons (never overrides tender hard limits)
  • Memory — Cross-work hints—always re-verify critical constraints

​
Phases (what “ready” feels like)

  • Empty — Search opportunities, browse catalog, research—matching still needs a file
  • Has file — Ask Orbid to analyze / scan the package
  • Scanned — Match a product, eligibility pre-check, risk flags
  • Matched — Deviations, go/no-go, fill Excel, generate packs
  • Exported — Review bid, record outcome
These map to product phase guidance (pills), not a rigid wizard you must click through.

​
Tools (optional detail)

Under the hood, outcomes map to tools such as tender scan, product match, Excel fill, go/no-go, and deliverable generation. Daily use: speak in outcomes. Specialists: Tools reference.

​
When Orbid should wait on you

  • Product selection / lock
  • Hard-limit failures and no-bid decisions
  • Weak evidence / residuals
  • Commercial pricing
  • Final package approval and submission
See Trust and judgment.