Knowl
Knowl is a local-first, governed memory system for AI coding agents. It stores typed, checkable facts with reasoning, evidence, provenance and supersession history so agents and teams get the current answer rather than stale notes.
At a Glance
- Individual developers using AI coding agents
- Software engineering teams and organizations sharing project decisions
- Teams running multiple coding-agent hosts or parallel worktrees
- CI and cloud-agent environments that need team memory without a local replica
- +1 more
AI Tools by Knowl
(1)
Knowl
Persistent Memory Layer for AI Agents
Discussions
No discussions yet
Be the first to start a discussion about Knowl
Latest News
Knowl 5.15.0: budget-composed session cards, separate subagent recall reporting and dated drift reporting.
Knowl 5.14.0: drift-scheduled turn-start cards, lexical transcript catch-up and fewer unnecessary hook processes.
Knowl 5.13.0: recall-gap measurement, re-derivation protection, category correction and optional local embeddings.
Knowl 5.12.0: relevance-floor wording narrowed to off-subject detection and transcript similarity made visible.
Products & Services
An installable Node 22+ binary (`@dat999zx/knowl`) that initializes a repository, maintains a local SQLite ledger and serves agent memory through MCP and CLI commands.
The hosted team layer where deliberately published facts are shared across workspaces; agents read a local replica, while publishing is explicit and server-side secret-scanned.
An experimental remote MCP endpoint for browser-based Claude and ChatGPT, and for agents/CI environments that cannot run a local replica. It supports workspace-scoped search, contradiction listing, recording and correction subject to granted access.
Market Position
Knowl positions itself against append-only agent-memory and project-context tools such as agentmemory, Mem0, Cognee, Letta/MemGPT and Zep by centering on freshness and explicit supersession: stale facts are retired from normal retrieval while remaining in history. It also differentiates through local-first storage, typed/provenance-rich atoms, MCP/CLI integrations and a publish-only-when-intended cloud layer. Its repository cites a 90% result on MemoryAgentBench FactConsolidation-SH at 262K context, while noting the benchmark's method and limitations.
Leadership
Founders
dat999zx
Independent maker and GitHub maintainer of the public dat999zx/knowl repository; the GitHub profile describes the creator as “Too human for AI,” and Product Hunt lists dat999zx as the maker. A personal legal name or prior-company background was not identified in the reviewed sources.
Founding Story
The maker says Knowl began with the recurring annoyance that an agent could work out a library choice, rate limit or rejected approach, then lose it when the session ended. The initial vision was persistent memory that is safer than append-only notes: every fact keeps its reasoning and evidence, and a changed fact supersedes the old one instead of leaving two competing answers active.
Business Model
Revenue Model
Freemium subscription for Knowl Cloud organizations, metered by newly published facts rather than reads; local CLI/engine use is free and open source, and queries are unmetered because agents read local replicas.
Pricing Tiers
One organization and workspace; 350 published facts/month (1,000 in the first month), 5,000 stored facts, unlimited members and 30 days of history.
2,500 published facts/month, 50,000 stored facts, 3 workspaces, unlimited members, 3 months of history and teammate-contradiction settlement.
15,000 published facts/month, 300,000 stored facts, 10 workspaces, unlimited members, 1 year of history, multiple organizations and everything in Plus.
Negotiated volume, unlimited history, SSO and SCIM, dedicated instance and everything in Pro.
Target Markets
- Individual developers using AI coding agents
- Software engineering teams and organizations sharing project decisions
- Teams running multiple coding-agent hosts or parallel worktrees
- CI and cloud-agent environments that need team memory without a local replica
- Organizations requiring SSO, SCIM, dedicated instances or controlled workspace access
- Persistent memory across Claude Code, Codex, Cursor and other AI coding-agent sessions
- Keeping architecture decisions, constraints, API facts and rejected approaches current as projects change
- Sharing verified project knowledge across teammates and repositories
- Running multiple agents or git worktrees against one shared project memory
- Using team memory from claude.ai, chatgpt.com or CI/cloud agents through a remote MCP connector