Internal AI Tools

Use a Rightbrain Task for one bounded internal operation, or an agent when the work requires runtime decisions, multiple tools, or conversation state. Deliver the same versioned capability through your API, triggers, or Connections to the systems where the team already works.

Why Tasks Work for Internal Tools

• Stable contract - Typed input and output keep downstream workflow code predictable.
• Reusable execution - The same Task can serve direct API calls, webhook-triggered work, and agent tool calls.
• Controlled context - Use Skills for reusable procedures and Collections for retrieved source material.
• Safe iteration - Revisions, evals, and rollback separate configuration changes from release decisions.

Common Internal Tool Patterns

Real-World Example: Legal Team Contract Analyser

The Challenge: Legal team reviewing 50+ vendor contracts monthly, each taking 45+ minutes to analyse for key terms, risks, and non-standard clauses.

Task Configuration:

• Input: Contract PDF upload
• Context: Company’s standard terms, red flag clauses, approval thresholds
• Output: Structured JSON with:
  • Contract type and parties
  • Payment terms and amounts
  • Key obligations for both sides
  • Identified risks with severity
  • Non-standard clauses highlighted
  • Recommended approval level

Deployment:

• Primary: Slack bot in #legal-contracts channel
• Secondary: API integration with contract management system
• Fallback: Direct API execution for ad-hoc reviews

How Legal Team Uses It:

1. Procurement drops contract in Slack
2. Legal team member clicks “Analyse Contract”
3. Task returns structured analysis in ~30 seconds
4. Team reviews flagged items, approves or requests changes
5. Analysis automatically logged in contract system

Why It Works:

• Configurable: Updated monthly as approval policies change
• Predictable: Always returns same structured format for downstream logging
• Stateless: Each contract analysed independently, no cross-contamination
• Composable: Same Task used in Slack and contract system API

Real-World Example: Sales-Call Deal Intelligence

The Challenge: Reps finish discovery calls with an hour of recording and no time to write it up, so budget, authority, and timeline detail get lost.

Task Configuration:

• Input: Call recording (WAV, MP3, and other common formats)
• Processing: An audio-transcription input processor turns the recording into text before the reasoning model sees it — the audio itself never reaches your analysis model
• Output: Structured JSON with the full BANT breakdown — budget, decision-makers, timeline, and the pain points named on the call

How It Runs:

One model transcribes the audio and a second reasons over the transcript, so you can pair the best transcription with the best analysis. On a real discovery call the transcription took about 10.8 seconds and the analysis about 35 seconds, extracting the budget, the named decision-makers, and an end-of-quarter timeline into a record your CRM can ingest directly.

Next Steps