Decision memory for teams and their AI agents
Precedent solves three problems: the reasoning behind a decision is lost, the decision never reaches the work it governs, and the next choice contradicts it without anyone noticing.
Free for up to 5 people. No card required.
Three problems, and what does the work
Memory, enforcement, and conflict. Each one has a feature behind it, and none of them ends with software deciding anything: every path finishes with a person reading the cited decision and confirming, revising, or superseding it.
The decision and the reasoning outlive the thread
It was settled in a channel six months ago. The thread scrolled away, the two people who argued it out have moved teams, and the only thing left is the code.
- Slack and Teams capture: a thread becomes a draft with the title, summary, rationale, and alternatives pulled out. If the model is down, the raw thread is still saved.
- Suggested drafts from linked channels, so decisions get recorded without anyone remembering to file one.
- Every record carries the rationale and the alternatives that lost, not only the outcome.
- Full revision history. Nothing is overwritten, and search runs over every revision.
- Hybrid search: keyword and embeddings together, so the database decision surfaces when you search for the storage choice.
- Evidence links to Jira, Linear, GitHub, Notion, and Drive, each with the sentence that supports the decision.
- Templates for architecture, vendor, pricing, and launch decisions so the record is complete on the first pass.
- Decisions attach to projects and domains, so you can read what constrains a piece of work before you plan it.
- Ask AI answers from the record and cites the refs behind every claim, using your own model key.
- Insights on the record itself: coverage per area, where decisions enter from, and which records have no evidence at all.
The record reaches the work, not just the wiki
A written decision that nobody reads at the moment of the change is a document, not a control. Agents make a dozen unstated choices per task and never look anything up.
- An MCP server your agents connect to, with search, node context, impact tracing, and drafting tools.
- The decision-librarian skill: search before acting, cite conflicts, answer why from the record instead of from memory.
- Harness hooks that run outside the model's discretion. One injects matching decisions into every prompt, one blocks the finish or the push with the conflicting refs.
- A GitHub Actions check that searches the record with the pull request diff, then comments the matching refs or fails the build.
- Read-scoped OAuth grants for CI and agents, with their calls recorded like anyone else's.
- Agents propose, people confirm. No tool can confirm, supersede, or archive a decision.
- Review dates and structured conditions on milestones, metrics, and dependencies. When one fires, a review lands in the owner's inbox.
- Per-node Slack digests, so a team sees what was decided and what is waiting on them where they already work.
- Every confirmation, revision, supersession, and review is a recorded event, exportable with a cryptographic signature.
Contradictions get raised and resolved, not buried
The new choice quietly contradicts a decided one. Nobody notices until production, and the old rationale is never revisited, so the same argument runs again next quarter.
- A search before the work returns the conflicting decision by ref, and the rule is to cite it rather than route around it.
- Supersession with a written reason. The predecessor flips to superseded and links forward to what replaced it.
- Only a decided decision can be superseded, and supersession links cannot be hand-edited or deleted afterwards.
- Impact tracing: everything that depends on or is affected by a decision, three links deep, with owners named.
- Dependency conditions open a review for the owners downstream when the decision they were built on changes.
- depends_on and affects links, so a decision sits in the graph it constrains rather than on its own.
- Conflicting and constraining decisions surface in node context before planning starts, not at review time.
- Course-change insights: confirmations against supersessions and reversals by month, so a stale record is visible.
Pricing
Start free and stay free while you are small. Pay when the whole team is in.
Free
The whole product, for a team small enough to fit around one table.
- Up to 5 members
- Decision log with full revision history
- Slack and Teams capture
- Ask AI with your own model key
- MCP access for your agents
- Evidence links to Jira, Linear, GitHub, Notion, and Drive
- Review dates and reminders
Teams
The same product without the seat limit, for when the whole team is in.
- Everything in Free
- Unlimited members
- The first 5 members stay free; $5 a month for each one after that
- Prorated as people join and leave
- Cancel any time, keep the record
Enterprise
For organisations that answer to auditors, regulators, or their own lawyers.
- Everything in Teams
- SAML and OIDC single sign-on
- Signed audit exports
- Legal hold
- EU data residency
- Priority support
Every feature works on both plans. Free stops at five seats, counting invitations you have sent but nobody has accepted yet. Upgrade under Settings, Billing when you need the sixth, and pay only for the members past the fifth.
Talk to us
Questions about Enterprise, data residency, or anything on this page? Email us and a person answers.