Get started
Pricing v1 is live · Free 1,000 contexts

Your agents learn all day. Let them sleep on it.

ContextQ gives AI coding agents persistent, searchable memory that compounds across conversations, projects, and tools. A dream pass clusters the day's scattered entries, writes durable summaries, and archives the originals.

Start free — no card See how it works
Works with the agents you already use Claude CodeCursorCodexGemini CLIWindsurfCline
How it works

Awake, the agent saves. Asleep, ContextQ keeps the gist.

In 1994, Wilson and McNaughton found that hippocampal cells which fired together while a rat explored fired together again in the sleep that followed, and not in the sleep before. ContextQ runs the same schedule on your agents' memory: sessions write entries, a dream pass replays what belongs together into one durable summary, and the next session searches the summary instead of the pile.

Fig. 1Hypnogram of one agent day, scored in the lab's order: W, R, N1, N2, N3.
a

Save a decision once

RUN · awake · W

# Save a decision once, available everywhere
ctx_save "Auth flow uses Zitadel JWT, RS256 only"
saved · ctx_4f2 · type=reference

Any agent writes through the same 16 MCP tools behind one endpoint. Entries touched in the same work fire together, like place cells on one track.

b

Dream: replay, cluster, keep the gist

POST · asleep · N2 → N3

# Synthesize across stored knowledge
POST /api/dream { "workspace": "acme" }
clustering 47 entries...
3 MOCs · 12 supersedes · 4 archived

An LLM-assisted pass clusters related entries, writes a map of content, and archives the originals. Typed links record it: supersedes, contradicts, derived_from.

c

Search the next morning

Lights on · next session · W

full-texttsvector + BM25
vectorpgvector cosine
Reciprocal Rank Fusionrank 1 · ctx_4f2
# Hybrid retrieval the next session, the next project
ctx_search "how do we sign tokens"
full-text + vector + RRF fusion
3 results · 92ms · rank=ctx_4f2

A different session, project or tool asks a vague question and gets the decision back. Decisions compound instead of evaporating.

Scoring key · what the marks on the record mean

Scoring key, part one
SOSlow oscillation, 0.5–2 Hz, at least 75 µV frontal. N3 is scored when it fills 20% of an epoch.the dream pass window
SPSpindle, 11–16 Hz, at least 0.5 s, maximal over central leads. Nests in the SO up-state.clustering related entries
RPLHippocampal ripple, about 80–120 Hz in humans, clustered in spindle troughs.one entry replayed
Scoring key, part two
MOCGist-like representation integrated in neocortex after repeated replay.durable summary, type=synthesis
ARCDown-selection: net synaptic strength renormalised during sleep.original archived, supersedes
WWake. Scored per 30 s epoch from lights out to lights on.sessions: save, search

Wilson & McNaughton, Reactivation of hippocampal ensemble memories during sleep, Science 1994 · Staresina et al., Hierarchical nesting of slow oscillations, spindles and ripples in the human hippocampus during sleep, Nat. Neurosci. 2015 · Klinzing, Niethard & Born, Mechanisms of systems memory consolidation during sleep, Nat. Neurosci. 2019 · Tononi & Cirelli, Sleep and the price of plasticity, Neuron 2014 · AASM Manual for the Scoring of Sleep and Associated Events.

Start free — no card Read the docs
Features

Memory that thinks, not just stores.

Vector DBs give you embeddings. ContextQ gives you a second brain that clusters, links, and prunes itself. Built on Postgres + pgvector + Elasticsearch with PKM-grade memory evolution.

Fig. 2Five properties of the record, one per cell.
a

Dream synthesis

Signature

LLM-assisted passes cluster related entries, write durable summaries, and archive the originals — keeping your memory atomic and discoverable. Think weekly retrospective for your second brain. Free runs one manual dream every 14 days; paid plans run it on their own.

  • deploy hits Lavender
  • use EasyPanel
  • SSH workaround
  • migrations first

MOC · deploytype=synthesis

Lavender prod via EasyPanel, SSH fallback when deployService fails, always migrate before promote.

b

Hybrid search

Full-text (Postgres tsvector + Elasticsearch BM25) fused with pgvector cosine via Reciprocal Rank Fusion. Find context fast, even with vague queries. Team adds cross-encoder reranking; Enterprise adds LLM query rewrite.

score = Σ 1 / (k + rank)

c

Typed links

[[auth-flow]] → supersedes

Wikilinks resolve to typed edges (supersedes, contradicts, derived_from). On paid plans, memory evolution auto-archives outdated entries.

d

16 MCP tools, one endpoint

v1

Drop the HTTP MCP URL + your API key into Claude Code, Cursor, Codex, Gemini, Windsurf, Cline. Every agent gets the same memory.

ctx_savectx_searchctx_getctx_listagent_bootagent_checkpointagent_handoffagent_handoff_save+8 more
e

Tamper-evident audit chain

Every write is hash-chained. Verify integrity on demand, export CSV, set per-tenant retention. Tenant-isolated row-level security.

sha256 7a3f…b2e9 → c14d…91a3 → verified

How we compare

Not just a vector DB. Not just a memory API.

ContextQ is the only stack that combines hybrid retrieval, PKM-grade evolution, and developer-tool ergonomics.

Table 1Capabilities by product.
Capability ContextQ Mem0 Zep Pinecone
Hybrid full-text + vector + RRF yes partial partial —
Dream / synthesis (LLM-assisted) yes Free: 1 manual / 14 days · Pro+: unlimited + auto — — —
Typed links + MOC clustering yes Pro+ — — —
Memory evolution (auto-archive) yes Pro+ partial — —
MCP native (16 hosted tools) yes — — —
Tamper-evident audit chain yes — — —
Bi-temporal lifecycle yes — — —
Entry-level price Free / $19 $19+ $99+ $70+

Plan tags mark ContextQ capabilities that need an LLM and a paid plan; everything untagged is on every plan, Free included. Comparison reflects publicly available pricing and feature pages as of May 2026. Independent assessment, not affiliated.

Get started

Three steps. Five minutes.

Two install paths — universal MCP, or the ContextQ plugin for Claude Code.

  1. 01

    Sign up and approve

    Sign up free, no credit card, and get a 14-day Pro trial. Run the Claude Code installer and approve once in your browser — no key to copy. Every plan can create dashboard API keys for other agents; Free includes 3 active keys.

  2. 02

    Connect your agent

    The plugin connects Claude Code on its own. For Cursor, Codex, Gemini CLI, Windsurf, or Cline, paste the MCP endpoint. Tools auto-discover.

  3. 03

    Save and search

    Your agents now share memory across sessions, projects, and editors. Decisions compound instead of evaporating.

Drop-in config.mcp.json

claude_desktop_config 
{
  "mcpServers": {
    "contextq": {
      "type": "http",
      "url": "https://app.contextq.dev/mcp",
      "headers": {
        "Authorization": "Bearer ctq_••••••••"
      }
    }
  }
}

Works the same in Cursor, Codex, Gemini, Windsurf, Cline.

Tool surface

16 tools. 4 groups. One memory layer.

Memory, session state, a shared goal board and workspace setup: the hosted MCP surface every agent shares. Dream passes and the rest of the platform run through the REST API.

Table 2The hosted MCP tool surface at app.contextq.dev/mcp, by group.16 tools
No.GroupToolsCount
01 Memory ctx_save · ctx_search · ctx_get · ctx_list 4
02 Session agent_boot · agent_checkpoint · agent_handoff · agent_handoff_save 4
03 Goal board goal_add · goal_advance · goal_frontier · goal_list · goal_sync_relay 5
04 Workspace ctx_workspace_list · ctx_workspace_ensure · ctx_project_ensure 3
Browse the full reference
Pricing

Start free. Scale when ready.

All plans include hybrid search, MCP integration, audit chain, and tenant isolation. Cancel anytime.

Free

For personal projects

$0 forever

  • 1,000 contexts
  • 2,000 code contexts
  • 5,000 queries / mo
  • 1 user
  • 3 active API keys
  • 1 manual dream / 14 days
  • CSV export
Get started

Most popular

Pro

For solo builders shipping fast

$19 /mo

  • 5,000 contexts
  • 5,000 code contexts
  • 20,000 queries / mo
  • 1 user
  • 20 active API keys
  • Unlimited dream, auto-triggered
  • CSV export
  • AI memory features
  • Webhooks
  • API key IP allowlist
Start 14-day Pro trial

Team

For teams scaling memory together

$79 /mo

  • 25,000 contexts
  • 25,000 code contexts
  • 150,000 queries / mo
  • 5 users
  • Unlimited API keys
  • Unlimited dream, auto-triggered
  • CSV export
  • AI memory features
  • Webhooks
  • API key IP allowlist
  • Search reranking
  • Team mode
  • Priority support
Upgrade to Team

Enterprise

For organizations with custom needs

Custom

  • Custom usage limits
  • Unlimited users
  • Unlimited API keys
  • Unlimited dream, auto-triggered
  • CSV export
  • AI memory features
  • Webhooks
  • API key IP allowlist
  • Search reranking
  • LLM query rewrite
  • Team mode
  • Priority support
  • SSO
  • Audit log export
Contact sales

All plans include tenant isolation and the tamper-evident audit chain.

AI memory features are the LLM-powered background work: atomicity and contradiction checks, memory evolution, skill distillation, memory extraction from conversations and ingest, conversation summaries, OCR for image-only PDFs, LLM handoff generation and automatic dream runs. Every plan keeps saving, listing, MCP access and hybrid keyword + vector search.

Every new account starts with a 14-day Pro trial, no card required. When it ends, the account moves to Free unless you upgrade.

Stop teaching your agents the same thing twice.

Free for your first 1,000 contexts. No credit card. Five minutes to set up.

Create your tenant Read the docs