Your AI agents
are forgetful.

Every agent starts cold. Primer gives Cursor, Claude Code, Windsurf, and 9 other tools persistent, trusted memory. They start informed, act safely, and hand off cleanly.

npm install -g getprimer

then cd your-project && primer init

Works withCursor · Claude Code · Claude.ai · VS Code · Antigravity · OpenCode · Windsurf · Zed · Gemini CLI · Claude Desktop

The problem

AI coding agents are powerful. They are also completely amnesiac.

Every session starts from zero

Your agent re-solves the same problem it fixed last Friday. It has no idea what you built yesterday.

Agents repeat the same mistakes

No persistent rules means no persistent learning. Every new chat brings the same bad patterns back.

Switching tools resets everything

Move from Cursor to Claude Code mid-feature. All context is gone. You brief from scratch.

No audit trail

You have no idea what your agent actually did, why it made a decision, or what it changed last Tuesday.

You brief agents, not build

Half your time goes to re-explaining your codebase. You are a context manager, not a developer.

Teams run isolated agents

Every engineer's agent starts cold. No shared memory, no shared rules. Coordination is impossible.

How it works

Persistent memory

Facts, rules, decisions captured automatically.

Every file you save, every agent session, every architectural choice — Primer records what matters and surfaces it to the next agent in any tool.

Facts, rules, observations and decisions in structured memory
Agents never start cold. They know your project before you say a word
Memory scoped per project. Shareable across repos with groups

Memory · 32 facts
All32
Pinned2
Facts11
Rules8
Observations9
Uses esbuild for bundling. Do not switch to webpack without testing all output targets.
fact2h ago
Never commit to main without running npm run build first.
rulepinned1d ago
Auth flow uses Supabase SSR. JWT stored in httpOnly cookie, not localStorage.
obs3h ago
Chose Supabase over PlanetScale. Row-level security was the deciding factor.
decision3d ago

Trust and conflicts

Catch when agents disagree. Resolve it once.

When two agents write conflicting facts, Primer flags the conflict. You resolve it once. Every agent sees the resolution going forward.

Conflicts detected automatically across all agents and sessions
Your decision is logged. Agents stop re-arguing the same point
Full history of what was said, by whom, and when

Conflicts · 1 open
Data fetching strategy
conflictcursor
Use React Server Components for data fetching
conflictclaude-code
Use client-side SWR hooks for all data fetching
resolvedyou
Decision: RSC for initial load, SWR for mutations

Guardrails

Define what agents can touch before they touch it.

Block writes to sensitive files, require review before migrations run, log every action. Policy enforces before the action, not after.

Rules defined in .primer/policy.json — version-controlled
Blocked actions are logged to the audit trail automatically
Agents are notified and continue with safe tasks

Handoffs · session 14
antigravitynext agent4h ago
Implemented project grouping. Added group field to .primer/config.json. getRemoteMemory() now filters by repo OR group. Migration applied.
Next: Add dashboard UI filter for group. Publish v0.14.11.

In practice

Your agent knows your project. Before you say a word.

exact output from v1.0.1

zsh — my-project

$ primer init

Step 1 of 3 — Sign in

✓ Signed in as you@email.com

Step 2 of 3 — Start memory daemon

✓ Daemon running on port 37373

Step 3 of 3 — Connect your AI tools

✓ Cursor configured

✓ Claude Code configured

✓ Windsurf configured

──────────────────────────────────────────────────

✓ Primer is ready. Project: my-project

──────────────────────────────────────────────────

# Open Cursor, start a new chat

> "What do you know about this project?"

Agent: "This is a TypeScript project using esbuild.

I can see a daemon in src/daemon/ and CLI in src/cli/.

You have 47 memory facts from previous sessions.

Want me to continue where you left off?"

Details

Built for how you actually work.

1.0 Memory

Structured memory. Not raw text.

Primer categorizes everything: facts (ground truth), rules (constraints), observations (patterns), and decisions (why you chose X). Agents query by type so a rule about committing is never confused with a fact about your database.

factArchitectural truth. Uses Postgres, not MySQL.
ruleConstraints. Never delete migrations.
obsPatterns noticed. Tests always fail on port 3001.
decisionRecorded choices. Chose esbuild over webpack.

Memory · 32 facts
All32
Pinned2
Facts11
Rules8
Observations9
Uses esbuild for bundling. Do not switch to webpack without testing all output targets.
fact2h ago
Never commit to main without running npm run build first.
rulepinned1d ago
Auth flow uses Supabase SSR. JWT stored in httpOnly cookie, not localStorage.
obs3h ago
Chose Supabase over PlanetScale. Row-level security was the deciding factor.
decision3d ago

2.0 Trust

Conflict detection across all your agents.

Two agents writing conflicting facts is caught automatically. You resolve it once — your decision is logged, and every future agent sees the resolution. Primer keeps the shared picture of your project consistent.

Conflicts · 1 open
Data fetching strategy
conflictcursor
Use React Server Components for data fetching
conflictclaude-code
Use client-side SWR hooks for all data fetching
resolvedyou
Decision: RSC for initial load, SWR for mutations

3.0 Guardrails

Policy that runs before the action.

Block writes to sensitive files, require human review for migrations, log every agent action. Rules defined in .primer/policy.json — plain JSON, version-controlled, reviewed like code.

Guardrails currently enforce at the MCP tool level in Cursor. Expanding to other tools iteratively.

Handoffs · session 14
antigravitynext agent4h ago
Implemented project grouping. Added group field to .primer/config.json. getRemoteMemory() now filters by repo OR group. Migration applied.
Next: Add dashboard UI filter for group. Publish v0.14.11.

Early users

From developers who used it first.

I used to re-explain my entire codebase at the start of every Cursor session. Primer just handles that now. It gets better the longer you use it.

Jabez Developer

The guardrails alone are worth it. My agent tried to delete a migration file. Primer blocked it before I even noticed. Saved me an hour of debugging.

Sam Developer

I switched from Cursor to Claude Code halfway through a feature. The agent picked up exactly where I left off. Same context, same rules. It felt like cheating.

Cyrin Developer

Our whole team uses it now. No one has to explain the codebase to their agent from scratch anymore.

Charls Developer

Setup took two minutes. I ran primer init, restarted Cursor, asked what it knew about the project. It described it perfectly.

Baruch M Developer

Your AI agents
are forgetful.

AI coding agents are powerful. They are also completely amnesiac.

Facts, rules, decisions captured automatically.

Catch when agents disagree. Resolve it once.

Define what agents can touch before they touch it.

Your agent knows your project. Before you say a word.

Works with the tools you already use.

Built for how you actually work.

Structured memory. Not raw text.

Conflict detection across all your agents.

Policy that runs before the action.

From developers who used it first.

Free to start.

Common questions.

Give your agents a memory.

Your AI agentsare forgetful.

AI coding agents are powerful. They are also completely amnesiac.

Facts, rules, decisions captured automatically.

Catch when agents disagree. Resolve it once.

Define what agents can touch before they touch it.

Your agent knows your project. Before you say a word.

Works with the tools you already use.

Built for how you actually work.

Structured memory. Not raw text.

Conflict detection across all your agents.

Policy that runs before the action.

From developers who used it first.

Free to start.

Common questions.

Give your agents a memory.

Your AI agents
are forgetful.