seamless

fair questions, longer answers

Seamless vs. the alternatives

Seamless is a local-first coordination substrate for the coding agents you run: persistent memory, dependency-aware tasks, and shared plans, stored as markdown files you own and indexed by one local Go daemon. This page compares it with the tools people actually weigh it against - Claude Code's Agent Teams, claude-mem, Anthropic's built-in memory, Mem0 and Zep, Dosu, agentmemory, and Supermemory.

Several of these are excellent, and for plenty of people the right answer is one of them; every section below ends with exactly when that is. What you will not find here is an accuracy benchmark. Memory benchmarks in this category are self-refereed, every vendor publishes one where they win, and we don't play. Legibility is the benchmark: open the folder, read what your agents wrote.

at a glance

Seven axes, no adjectives.

tool storage format where it runs account required multi-agent coordination forgetting policy runtime cost
Seamless markdown files in git; SQLite indexes them and stores operational state your machine, localhost only none ready-queue, lease claiming, shared plans proposals only - you approve Go daemon + CLI, no CGO free, MIT
Agent Teams team task list inside Claude Code inside Claude Code a Claude subscription shared task list, claims, dependencies - one team state ends with the team built into Claude Code included
claude-mem SQLite + Chroma vectors under ~/.claude-mem your machine; optional cloud sync none none - memory only none documented Node 20 + Bun free, Apache-2.0
Anthropic memory files in a managed sandbox Anthropic's sandbox a Claude account none - one agent's silo model-managed edits built into Claude included
Mem0 / Zep vector store / temporal knowledge graph cloud-first; self-host available yes none - a memory API automatic extraction and consolidation SaaS or self-hosted services free tiers, paid plans
Dosu hosted knowledge base cloud yes - team shared team knowledge, no task claiming managed for you SaaS commercial
agentmemory SQLite + vector index; its DB is the truth your machine; cloud optional none shared memory namespaces automatic - decay, TTL, eviction Node 20 + iii-engine free, Apache-2.0
Supermemory cloud store; a local drop-in exists cloud-first yes, for the full product none - a memory API automatic forgetting SaaS; small Go drop-in free tier; the Claude Code path needs Pro

# facts checked against each project's public docs, July 2026. Got yours wrong? Open an issue - we'll fix it.

the field

Every close alternative, taken seriously.

What each one is, where the line between us actually sits, and when to pick it instead of us.

Claude Code's Agent Teams

first-party multi-agent coordination

Agent Teams are Claude Code's own answer to parallel agents, and it is a real one: a shared task list, task claiming backed by file locking so two agents cannot grab the same step, addBlockedBy and addBlocks dependency markers that gate what is claimable, and git worktrees isolating each agent's edits. If that list sounds familiar, it should - it is the shape of our ready-queue and lease claiming, shipped first-party and free. We would be lying if we called it anything but strong.

The line is lifetime and reach. An Agent Teams task list belongs to one team, inside one tool, for the life of that team: when the team winds down, the claims, the ordering, and the context go with it. Nothing crosses to your next session, to another checkout, or to an agent that is not Claude. Seamless is the layer underneath: a claim outlives the team, a plan outlives the week, and a Codex agent can read what a Claude agent wrote - because all of it is markdown and SQLite on your disk, in git, not state inside one product.

In practice they compose. Spin up an Agent Team for the burst of parallel work; let Seamless hold the plan, the constraints, and every claim that has to survive it.

choose them instead if

Everything you need coordinated happens inside one Claude Code team and nothing has to outlive it. It is first-party, free, and needs no install.

claude-mem

transcript memory for Claude Code

claude-mem is by far the most popular memory tool in the Claude Code ecosystem, and deservedly: install the plugin and it just works. Hooks watch your session, an AI pass compresses what happened into semantic summaries, and future sessions start with that context loaded. Everything stays local - SQLite plus a Chroma vector store under ~/.claude-mem - it is Apache-2.0, free, and reaches beyond Claude Code to OpenCode, Copilot, Gemini, and friends.

The difference is what gets remembered, and who decides. claude-mem records what happened: its store is generated summaries of your sessions, kept at the compressor's judgment. Seamless stores what an agent - or you - deliberately wrote down: a constraint, a runbook, a decision, in an eight-kind taxonomy with explicit supersession and a gardener that proposes cleanup you approve. One is a memory of events; the other is a curated body of knowledge you can open in an editor and diff in git.

And memory is claude-mem's whole surface. There is no task queue, no claiming, no plans - it will not stop two agents from doing the same work twice.

choose them instead if

You want effortless recall of what past sessions did, today, with the biggest community in the space - and you don't need shared tasks or a store you author yourself.

Anthropic's built-in memory

first-party per-agent memory

Claude's managed memory gives an agent a directory it can read and write across sessions, mounted into a sandbox and managed for you. It is first-party, zero-setup, and improving steadily - for a single Claude surface it may honestly be all you need.

It is also, by design, a silo. The directory belongs to one agent in one product: no cross-agent sharing, no cross-client reach, no tasks, no claiming, no plans. And the files live in Anthropic's sandbox rather than on your disk - you cannot grep them, diff them, or commit them alongside the code they describe. Seamless is the shared ground every agent you run stands on, whatever client it speaks from, and its files answer to ls.

choose them instead if

You use exactly one Claude surface, can't or won't install anything, and continuity for that single agent is the whole requirement.

Mem0 and Zep

memory APIs for products you build

Mem0 and Zep are serious infrastructure: SDKs and APIs that give the product you are building per-user memory. Mem0 extracts facts from conversations into a vector store; Zep builds a temporal knowledge graph that can answer questions like who owned this account in February. Both are production-oriented, well-funded, and cloud-first, with self-hosted open-source paths.

They solve a different problem than Seamless does. Their customer is a developer shipping an app whose users need memory at scale; the memory belongs to your product and lives behind an account. Seamless is for the agents you run - the fleet of coding sessions pointed at your own repos - on your machine, bound to localhost, with no account anywhere. Comparing them axis by axis mostly shows the axes mattering differently: multi-tenancy and SLAs on their side, git-diffability and task claiming on ours.

choose them instead if

You are building a product whose users need memory. That is what they are for, and it is not what Seamless is.

Dosu

team knowledge infrastructure

Dosu is knowledge infrastructure for whole teams: a hosted, shared brain that captures knowledge as your agents and engineers work, answers questions with it, and keeps your documentation in sync with the code. For an organization, that docs-that-heal loop is genuinely hard to replicate, and Dosu does it well.

Seamless is the local-first counterpart, scoped to one person and their fleet. Durable knowledge lives in a folder in git on your own disk rather than in a vendor's cloud, and there is no Seamless account. Model calls leave the machine only when you configure a remote provider; Ollama keeps them local. Seamless also adds what a team-knowledge product does not try to: coordination primitives - a dependency-aware ready-queue, lease-based claiming, shared plans - for the agents actually doing the work.

choose them instead if

A whole team needs shared, managed knowledge with documentation kept in sync, and having a vendor run it is a feature rather than a concern.

agentmemory

the feature-maximal local memory runtime

agentmemory is the most feature-dense local memory system in the field, and on raw memory features it out-ships us: triple-hybrid retrieval (BM25 plus vectors plus a graph), a dozen ambient hooks, 53 MCP tools, contradiction detection, tiered consolidation, session replay, and benchmark results it publishes proudly. It runs locally on SQLite, embeds offline for free, and is Apache-2.0.

The disagreement is philosophical, and worth stating plainly. agentmemory automates the lifecycle: Ebbinghaus-style decay, TTLs, and importance eviction decide what fades, and its database is the source of truth, with Obsidian export as a secondary view. Seamless holds the opposite bet: markdown files are the truth for durable knowledge, forgetting is a proposal you approve, and the whole system stays small enough to read - two small Go binaries against a Node 20 runtime plus its separate engine. Whether 53 tools is a feature or a smell is exactly the question to ask yourself.

choose them instead if

You want the most automated, most feature-complete local memory available, and you are comfortable letting a decay curve do the forgetting.

Supermemory

a memory API for your whole digital life

Supermemory positions itself as the memory API for the AI era, and its scope is the widest here: a personal-and-company brain with connectors for Google Drive, Gmail, and Notion, multimodal ingestion, and fast retrieval, run as a cloud service. An open-source local drop-in exists - a small static Go server that runs offline with Ollama - but the cloud is the product.

That breadth is the trade. Seamless is narrow on purpose: coding agents, their shared knowledge, and their division of labor, local-first with no account. Supermemory's Claude Code integration also sits behind its Pro subscription, while Seamless is free everywhere it runs. If your question is what should remember my email and my meeting notes, that was never our question.

choose them instead if

You want one memory across your tools and your life - email, docs, notes, agents - and a cloud service holding it is acceptable.

see for yourself

The folder is the argument.

Install it, run two sessions, then open ~/.seamless/memory and read what got written.