For security and platform engineering teams
Decision governance that your engineers and their AI agents follow.
Precedent records what your team has agreed: choices, rules, exceptions, how-tos, facts, lessons and terms, each with its reason, its approvals, its evidence and the date it ends or must be reviewed. People and AI assistants read them before they act in that Area, and a pull request check fails while an exception in the Areas it checks has expired.
The seven TypesFree is for individuals and needs no credit card. It connects to Claude, Claude Code, Codex, Cursor and VS Code.
Fictional walkthrough · Acme Robotics
A cheaper quote needs a second look
See how a recorded reason changes the next conversation, whether it happens with a colleague or a coding agent.
The request
“Moneyman has the cheapest quote. Let's use it for checkout.”
The price is visible. The earlier evaluation is easy to miss.
The recorded reason
AR-112DecidedDo not use Moneyman while the UAE is a launch market.
Moneyman has the cheapest quote, but it cannot settle in the UAE. The launch needs UAE settlement.
Review when Moneyman supports UAE settlement.
The informed response
“AR-112 records why Moneyman was rejected: it cannot settle in the UAE. Have the launch requirements changed?”
The team can check the reasoning before repeating the evaluation.
Illustrative response from a connected assistant. It retrieves decisions you can access; you review their relevance.
How assistants use your decisionsExceptions that end, approvals set per Area, and a check where the code changes
Security and platform teams answer the same questions in every audit: who approved, who asked for changes, which waivers are open, and whether the work followed the standard. Decision governance in Precedent records the answers on the decision itself and surfaces them where engineers work.
Waivers that end on a date
An exception names the standard it departs from, the compensating controls and the day it ends. From the next day it shows as Expired, the standard applies again, and the owner was reminded 30 and 7 days before.
Record an exceptionApprovals
Each Area sets how many people must approve a decision, and whether its owner's approval counts.
Approve decisionsA check on every pull request
A GitHub Action comments with the decisions that apply to the Areas a pull request touches. It fails only on recorded facts: an expired exception in a checked Area, or a superseded decision the pull request cites.
Add the checkEvidence for your auditors, not a second GRC platform
The approval trail, exception expiry, revision history and signed audit export map to SOC 2 and ISO 27001 control families. Precedent records engineering decisions; it is not a risk register.
What Precedent evidencesAgents follow the decisions your team governs. People stay in charge.
Connect your assistant once. It reads the decisions for the Area it is working in and is warned before it acts when an exception has expired, a standard was superseded or a review is overdue. It can propose a decision or an exception and mark a condition as met. To approve, revise, supersede or reverse a decision, it asks you to accept the change first.
Read what was decided
Claude, Claude Code, Codex, Cursor and any remote MCP client ask Precedent for current guidance in the Area they are working in, with warnings for expired exceptions, superseded decisions and overdue reviews. Drafts are shown as drafts, never as settled.
Connect your assistantPropose a draft for its owner
When a conversation reaches a choice, or needs an exception, the agent proposes a draft with its rationale and alternatives. It lands in the owner's inbox. Each approval records the person by name.
How approving worksSet up once per project
The plugin saves a credential-free .precedent.json per project naming where to record and up to three Areas to check. It needs no Python, npm package or local server.
Install the pluginAgents propose, and a person accepts each change.
A person with access approves or changes a decision, in the web app or by accepting the change in their agent, and, where the workspace allows it, can approve a draft or reaffirm a review from its post in Slack. Related results are a starting point for review, not a guarantee that every conflict has been found.
How review worksMost decisions are made in a thread, and Precedent turns that thread into a draft.
Type /precedent what was decided in a Slack thread, or use the “Draft a decision” message shortcut. The thread becomes a draft with a real title, summary, rationale, alternatives and review date or condition, and the thread itself attached as evidence. If it read more like an open discussion than a decision, the reply says so.
Link a channel to an Area and it works both ways: drafts made in the app and decided decisions are announced there, and decisions captured there are filed under the Area automatically.
See how capture worksDecisions that know when to be revisited
Every decision carries a review, on a date or when a condition is met. Give the condition a shape, such as a milestone, a dependency or a date, and Precedent opens a review for the owner when it is met, whether that happens by a person, an agent, an integration or the calendar. The decision’s status never changes by itself.
Reaffirm, reverse or supersede with the history intact. The old reason stays readable next to the new one.
See how to review a decisionBuilt so the record can be trusted
The same visibility rules apply to a colleague searching, an agent reading over MCP, and an export leaving the system.
- Permissions per Area. Groups hold Viewer, Member, Maintainer or Area administrator roles. Every list, search, agent tool and export applies the same visibility. An agent sees only what its person can see.
- Evidence that stays put. Link Jira, Linear, GitHub, Notion or Google Drive and the reference is frozen at capture. Upload PDFs and images served only to people who can see the decision.
- Seven Types for what a team agrees. Every decision has one of seven Types: choice, rule, exception, how-to, fact, lesson or term. The Type sets what the decision holds and how assistants receive it.
- Ask AI from the first day. It answers from Precedent's model, or from your own Anthropic, OpenAI or OpenRouter account once you connect one. It uses the same read tools as your agents and runs as the signed-in person.
Start on Free, then pay per member when your team joins.
Free covers one person with full agent access. Teams is $5 per member per month. Enterprise adds single sign-on, signed audit exports, legal hold and EU data residency.
Free
Keep the reasoning behind your own decisions, and let your agents read it.
- One person, one personal workspace
- 100 new decisions per month
- 100 AI calls per month
- Full revision history
- Ask AI, included or on your own model key
- MCP access for your agents
- Automatic conflict checks
- No card and no expiry
Teams
The whole product, for teams who want the reasoning to outlive the thread.
- 1,000 new decisions per workspace per month
- 500 AI calls per workspace per month
- Higher-capability AI models may be included in future
- Decision log with full revision history
- Slack capture
- Ask AI, included or on your own model key
- MCP access for your agents
- Evidence links to Jira, Linear, GitHub, Notion, and Drive
- Prorated as people join and leave
- Cancel any time, keep the record
Enterprise
For organisations with audit, access, and data residency requirements.
- Everything in Teams
- SAML and OIDC single sign-on
- Signed audit exports
- Legal hold
- EU data residency
- Priority support
Monthly allowances reset on the first day of each month (UTC). Free and Teams currently use the same automation model. Teams may offer a more capable model in future. Ask AI uses Precedent’s model until you connect your own model account, which your provider bills separately. Reaching a creation limit preserves your existing records.
Talk to us
Questions about Enterprise, data residency, or anything on this page? Email us and a person answers.
Start with one decision you keep explaining
Bring a choice your team has already made. You can start in the app without connecting a model or an assistant.
- Record the reason. Create a workspace and an Area, then write the choice, why it won, and what would make you reconsider. Writing guide
- Make it usable. Add the evidence, review the draft, and approve it. Approval guide
- Bring it into the next conversation. Connect your assistant and ask why the choice was made, or find the record in search. Connection guide
Free is for individuals and needs no credit card. It connects to Claude, Claude Code, Codex, Cursor and VS Code.