Native AI

Fishladder includes a built-in AI layer that works across Feedback, Projects, Roadmap, and Hiring. It can analyze data, suggest actions, detect patterns, and automate repetitive work — all under your control through a permission model you configure.

Getting Started

AI features require an Anthropic API key. A workspace admin configures this in the environment. Once set, the AI sparkle button (✨) appears in the header bar across all apps.

The AI Panel

Click the sparkle button in the header to open the AI Panel. It slides out from the right side and has three tabs:

• Activity — A chronological feed of every action the AI has taken or proposed. You can approve, reject, or undo actions directly from this list.
• Signals — Patterns the AI has detected across your data, such as recurring feedback themes, stalled tasks, or pipeline bottlenecks. Signals include a frequency count showing how often the pattern has been seen.
• Chat — A conversational interface where you can ask the AI questions about your workspace data, request summaries, or trigger actions. The AI streams its responses in real time.

Permission Model

The AI permission model gives you full control over what the AI can do autonomously. Go to Settings → AI Settings to configure it.

There are three permission levels:

• Act Freely — The AI executes actions automatically (e.g., creating tasks from transcripts, grouping feedback). All actions are still logged and can be undone.
• Ask Before Acting (default) — The AI proposes actions and waits for your approval before executing. Pending actions appear in the Activity tab with Approve/Reject buttons.
• Human Only — AI analysis and suggestions are disabled for this module. Existing data is unaffected.

You set a default permission level that applies to all modules, then optionally override it per module. For example, you might set Feedback to "Act Freely" but keep Hiring at "Ask Before Acting" for more sensitive decisions.

Module Toggles

Each AI module can be individually enabled or disabled:

• Feedback — Transcript analysis, auto-grouping, deduplication
• Projects — Task generation, assignment, priority, blocked detection
• Roadmap — Project suggestions from feedback, goal connections
• Hiring — Candidate scoring, pipeline bottleneck analysis
• Cross-Module — Feedback routing, signal analysis, follow-through checks

Action Approval & Undo

Every AI action is recorded with full context — what it did, why, and what data it used. From the Activity tab you can:

• Approve a pending action to execute it
• Reject a pending action to discard it
• Undo an applied action to revert it, with optional feedback about why

Undo feedback helps the AI learn from corrections. You can tag undo reasons (e.g., "wrong priority", "duplicate") to help improve future suggestions.

Signals

Signals are patterns the AI detects automatically as it processes data across modules:

• Recurring feedback themes — The same issue mentioned across multiple entries
• Stalled tasks — Tasks with no activity for an extended period
• Pipeline bottlenecks — Hiring stages where candidates get stuck
• Cross-module patterns — Connections between feedback, tasks, and roadmap items

Each signal tracks its frequency and when it was last seen. Resolved signals are archived automatically.

AI Chat

The Chat tab provides a conversational interface where you can:

• Ask questions about your project data, task status, or feedback trends
• Request summaries of recent activity or team workload
• Get explanations of AI-proposed actions
• Trigger AI workflows by describing what you need

Chat is rate-limited to 60 messages per hour per team. Responses stream in real time and your last 20 messages are saved between sessions.

Rate Limits

AI features have per-team rate limits to manage API usage:

• Chat — 60 requests per hour
• Feedback ingestion — 20 requests per hour
• Task generation — Standard API rate limits
• Candidate scoring — Standard API rate limits