Capture the why behind every decision

The rationale layer for every business workflow.

Captures the unwritten reasoning behind decisions in the apps your team already uses, Stripe, Jira, Gmail, Notion, and more.

Book a demo See it inline
Runs locally Self-hostable MCP-native
✦ Tacit Paradigm · capture rationale AI

What's the strategic reason behind this decision?

Tacit predicted the answer from your team's past decisions on similar actions.
Cross-train juniors before the lead's leave Capacity reallocation
Why we exist

Every decision logs the what. None of them log the why.

The reasoning lives in someone's head, and evaporates the moment they switch tabs.

76%

Operational decisions are tacit

Of the decisions logged in tools like Jira and Stripe, more than three quarters carry no recorded rationale beyond a status change or an amount field.

40h

Lost when one senior employee leaves

Estimated context lost per departing senior IC across the first 90 days post-exit, manifesting as duplicate work, repeated mistakes, and broken playbooks.

3-5×

Slower to reconstruct than to capture

Audit, compliance, and post-mortem teams routinely spend 3-5× the original decision time reverse-engineering reasoning that could have been captured in 30 seconds.

How Tacit works

Three layers, one platform.

Capture in the apps your team uses. Store as an institutional graph. Serve back to humans, agents, and models.

1

Capture

Native UI inside Stripe, Jira, Zendesk, Google Docs, Gmail, and any macOS app. Asks one selective question right when a real decision is being made.

  • Inline prompts that match each surface's design
  • Novelty-gated, silent on routine actions
  • Predicted answers as one-click chips
  • Free-form when chips don't fit
2

Graph

Every captured rationale becomes a node in a Document → Change → Rationale graph. Searchable, exportable, never lost when the person who made the call leaves.

  • Neo4j-backed institutional knowledge graph
  • Vector-indexed for semantic recall
  • Cross-surface, one ticket, multiple apps
  • Self-hostable in your VPC
3

Serve

Surfaces past rationales back when they matter, to humans through the dashboard, to AI agents through MCP, and into the supervised fine-tuning pipeline of any custom model your team trains.

  • Dashboard for review, audit, and onboarding
  • MCP server for any compatible AI agent
  • Supervised fine-tuning corpora, SFT-ready
  • API + SDK for in-product retrieval
Two ways to capture

Chrome extension. macOS app.

Together they cover every surface your team makes decisions in.

Chrome extension

Covers the browser surfaces: Stripe, Jira, Zendesk, Google Docs, Gmail, Notion. The prompt injects inline and adopts each host's design language, so it feels like it shipped with the app you're already using.

5+ first-party integrations MV3 Async + sync flows

macOS native app

Covers everything else: TextEdit, Notes, Slack, Linear, internal admin tools, vendor portals. Runs local models on-device, so screen content never leaves the laptop.

Any macOS app Runs local models Menu bar app
Integrations

Works with the tools your team uses.

First-party support for the surfaces where real decisions happen.

Stripe
Refunds, disputes, plan changes
Jira
Reassignments, status changes
Zendesk
Escalations, SLA exceptions
Google Docs
Strategy, contract redlines
Gmail
Customer-facing replies
Credit Union Centre
Loan-officer decisions
Slack
Cross-surface signal
Linear
Beta · Q3 2026
Notion
Roadmaps, specs, redlines
Salesforce
Coming soon
The prediction layer

Predicts the answer before it asks the question.

As your rationale graph grows, Tacit predicts the most likely answer for any new decision. Most prompts are answered with one tap on a chip, free-form is reserved for genuine novelty. Selectivity is the point: silence on routine actions, attention only when it matters.

71%
Chip acceptance rate (one-tap rationales)
138ms
Avg latency · predict_rationale call
87%
Of captures stay silent (similarity gate)
Why is this being closed as Won't Do?
✦ Resolved upstream · Safari 17.4 fix Cannot reproduce Owned by another team Out of scope
# How Tacit decided which chip to highlight if cosine(action, last_50_captures) > 0.85: → silent elif top_chip_confidence > 0.75: → chip-only elif decision != policy_default: → ask
For your AI stack

From decisions to better models.

Every captured rationale is a labelled (question, answer, context) pair, exactly the shape supervised fine-tuning wants.

MCP server · agent-ready

Standard Model Context Protocol. Plug Claude Code, Cursor, internal copilots, or any MCP-compatible client into your rationale graph in one line of config.

# claude_desktop_config.json { "mcpServers": { "tacit-paradigm": { "command": "tacit-mcp", "args": ["--workspace=acme"] } } }

Supervised fine-tuning · on your decisions

The captured graph is a clean SFT dataset by construction, every node is a real decision your team made, with the question, the chosen answer, and the chip set we surfaced. Train Llama, Qwen, GPT, or Claude variants on the way your team actually thinks.

# Build an SFT corpus from your rationale graph $ tacit sft build \ --surface=stripe,jira,gmail \ --format=instruction-tuning \ --out=./tacit-sft.jsonl # 4,872 (question, answer) pairs · 0 manual labelling # Pipe straight into Together / Anyscale / OpenAI SFT.
Customers

Built for the people making the calls.

Configured per role, each team sees the rationales relevant to their surface.

Finance Ops

Refunds, disputes, plan changes.

Store the tacit “avoid chargeback” tradeoffs refund teams already make, then surface precedent on the next similar case.

Example: Partial refund to stop a chargeback, reasoning replayed on the next six partials.
Engineering

Reassignments, Won’t Do, priority shifts.

Capture why tickets move in ways that look wrong on paper, before it disappears into history.

Example: P0 reassigned + points bumped, Tacit asks for the cross-training rationale.
Customer Success

Escalations, SLA breaks, retention.

Turn one-off CSM judgment into precedent for the next similar fire drill.

Example: $200 goodwill over a $100 cap after a long outage, proportionality saved for next time.
Sales

Counters, redlines, upgrades.

Make concessions and deal structure legible to the next AE on a similar account.

Example: Renewal counter + short discount, YoY optics strategy captured once.
Compliance & Audit

Living decision trail.

Query document → change → rationale in minutes instead of reconstructing mailboxes.

Example: All above-band SLA exceptions pulled in one graph query.
Lending & Underwriting

Manual calls on the file.

Tie officer judgment (amount, term, payoff) to the application ID for review and onboarding.

Example: Reduced loan + controlled payoff, rationale on RPL-2026-0418-0937.
Built for the enterprise

Privacy-first by design.

Inference runs locally. The graph runs where you want it to.

Runs locally

The macOS app runs local models on-device. Screen content never leaves the user's machine.

  • No cloud round-trip per capture
  • Works offline / air-gapped

Self-hostable

The backend, graph store, and MCP server can all be self-hosted in your environment.

  • Containerized · drop-in deployment
  • BYO LLM provider

Stop losing the why.

See Tacit running on your tools in a 30-minute demo.

Book a demo Read the docs