v0.6.1 · Now live

AI agents that
actually remember.

Persistent memory between sessions. Guardrails before dangerous writes. A full audit trail. One extension, every agent.

VS Code Marketplace · No account required to start

VS CodeCursorClaude.aiWindsurfAntigravity
15
MCP tools
0
Context lost per session
Agents, one memory
<1%
Token overhead
The problem

Same agent.
Completely different experience.

Most developers spend more time re-explaining context than they realize. Primer fixes that at the infrastructure level.

✕  Without Primer
Forgets everything daily

Re-reads your README, re-asks about architecture, re-discovers what was done. 15–30 tool calls before writing a line.

Edits files it shouldn't

No off-limits concept. Overwrites WIP, touches .env, blows away migration files.

No record of anything

Something broke. No idea what changed, in what order, or why. Every debug session starts blind.

✓  With Primer
Resumes exactly where it left off

One brief call delivers pending plans, past decisions, and handoff notes. Agents go straight to work — zero ramp-up.

Guardrails that actually stick

Dangerous files blocked before the write. Rules defined once, enforced across Claude, Cursor, Windsurf automatically.

Full audit trail, always

Every action signed and timestamped. See what changed, which agent did it, and when — in real-time and in history.

Features

Everything agents need
to work without supervision.

Primer runs as a local daemon alongside your IDE — zero latency, no cloud dependency for core features.

🧠
Persistent Memory

Facts, decisions, and handoffs survive session restarts. Agents on Machine B pick up exactly where Machine A left off.

🛡️
File Guardrails

Block, warn, or allow per file or pattern. Secrets are never touched. Rules sync across all agents instantly.

Live Activity Feed

Every agent action in real-time in your VS Code sidebar — file reads, writes, guardrail checks as they happen.

🔗
Multi-Agent Coordination

Conflict detection prevents two agents writing the same file simultaneously. Locks released automatically when done.

🔏
Cryptographic Audit Trail

Every action signed and timestamped locally. Tamper-evident logs mean you always know what happened.

📡
MCP Protocol Native

Works over stdio (Cursor, Claude Desktop) and HTTP/SSE (browser agents). One extension, all your agents.

How It Works

Three calls.
Full context. Zero repetition.

Agents call Primer tools instead of re-reading files. Context is always ready — preloaded, not rediscovered.

STEP 01
Start every session with a brief

One call loads all rules, past decisions, and handoffs. No re-reading README files.

primer_agent_brief({ session_id, agent })
STEP 02
Check before touching any file

Every write validated against guardrails. Blocked files rejected instantly.

primer_agent_check({ file_path, action })
STEP 03
Hand off to the next agent

Summary, next steps, and files stored. The next agent gets full context.

primer_agent_handoff({ summary, next_steps })
Connect an Agent

Works with every AI tool
you already use.

Primer exposes a local MCP server. Point your AI tool at it — that's all it takes.

⌨️ Cursor / Claude Desktop / Windsurf

One-click config injection via the VS Code sidebar.

1.Install Primer from VS Code Marketplace
2.Open Primer sidebar → Connect an Agent
3.Click ⊕ Add to Cursor — config written automatically
4.Restart your IDE. Tools appear instantly.
🌐 Claude.ai / Browser Agents

HTTP relay — copy the URL, paste into Claude.ai.

1.Install Primer + sign in with Google
2.Primer sidebar → + New URL → copy relay URL
3.Claude.ai → Settings → Integrations → paste URL
4.Claude uses Primer tools on its next message.

“The next agent always knows what the last one did.”

That's the whole point of Primer.

↓ Install Primer — it's freeOpen dashboard →