# Docmancer

> Work from the same local memory as every other coding agent on this machine. Recall prior decisions, preferences, instructions, and project conventions that Claude Code, Codex, Cursor, and other agents wrote here, with cited sources, fully local. Also searches a separate local technical-documentation index.

- **Type:** Skill
- **Install:** `agentstack add skill-docmancer-docmancer-docmancer`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [docmancer](https://agentstack.voostack.com/s/docmancer)
- **Installs:** 0
- **Category:** [AI & ML](https://agentstack.voostack.com/c/ai-and-ml)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [docmancer](https://github.com/docmancer)
- **Source:** https://github.com/docmancer/docmancer
- **Website:** https://docmancer.dev

## Install

```sh
agentstack add skill-docmancer-docmancer-docmancer
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# docmancer

An AI-agent memory harness: shared memory for coding agents.

Work from the same local memory as every other coding agent on this machine, instead of making the user explain the project again. Claude Code, Codex, Cursor, Gemini, and other coding agents write memory, instructions, and rules all over the machine, each locked in its own silo. Docmancer discovers that evidence, combines the durable parts into one human-readable Shared Memory tree, and gives every agent grounded, cited recall over it: automatically through recall hooks (Claude Code and Codex), on demand through this skill and the CLI, and through MCP tools for everything else. The core memory path runs locally. Network access happens only when the user explicitly fetches online documentation, uses an external model, checks a package registry, or enables Cloud.

## Quick start

```bash
pipx install docmancer
docmancer setup                              # discovers agents, indexes their memory, installs skills and hooks
docmancer ask "why did we pick Railway?"     # grounded answer with citations from what your agents wrote
```

If you use uv, `uv tool install docmancer` installs it the same way.

The default profile uses SQLite FTS5, sqlite-vec, and bundled Model2Vec embeddings: no API keys, no daemon, no model download, offline at runtime. The optional scale profile uses FastEmbed, sparse retrieval, and Qdrant without changing memory authority or provenance semantics. Technical documentation remains a separate Library corpus, searched with `docmancer docs`.

## Workflow

1. Run `docmancer ask ""` when prior decisions, instructions, conventions, or preferences may matter.
2. Follow stable citations with `docmancer read ` when the full file or current content hash is needed.
3. When the user explicitly asks to manage durable memory, use `docmancer ask` to prepare one complete-file proposal. Apply it only with the user's confirmation or explicit `--apply`.
4. Keep memory and technical-documentation results separate.
5. Use `docmancer docs list` and `docmancer docs query` for libraries, APIs, and vendor documentation.

Read-only Ask reads the latest committed local index. A configured generation provider is called by default after retrieval to produce grounded prose. Explicit mutation requests use one structured provider call to prepare one `create`, `edit`, `pin`, `move`, `duplicate`, `trash`, or `restore` proposal. Use `--read-only` to suppress action planning, `--apply` only after explicit authorisation, `--no-answer` for evidence only, and `--fresh` when the question must first wait for changed agent sources to be indexed.

## Memory commands

```bash
docmancer setup
docmancer web
docmancer ask "what deployment decisions apply?"
docmancer ask "how did this policy change?" --history
docmancer ask "return evidence only" --no-answer
docmancer ask "remember that production releases require a smoke test"
docmancer ask "update decisions/deployment.md to require two reviewers" --apply
docmancer common
docmancer delivery
docmancer timeline
docmancer memory canonical
docmancer read docmancer://memory/
docmancer write "# Decision\n\nDeploy on Railway." --path decisions/deployment.md --scope project
docmancer edit docmancer://memory/ - --expected-hash 
docmancer move docmancer://memory/ decisions/hosting.md --expected-hash 
docmancer import ./notes
docmancer status --check
```

Existing-file mutations require the current content hash. Conversational Ask proposals affect exactly one complete file and never apply because the user merely types “yes”. Imported and harvested sources remain read-only. Never trash, restore, connect Cloud, change capture installation, or publish Team memory without explicit user authorization.

When the user explicitly authorizes Personal Sync, `docmancer cloud connect` is the normal entry point. The first connected machine creates and checks the recovery kit and starts the first encrypted transfer. A second machine displays a four-word pairing code for approval from a connected machine; `docmancer cloud connect --recover` is the fallback when no connected machine remains. `docmancer cloud sync` is an explicit retry or later push and pull, not an onboarding requirement.

## Documentation commands

```bash
docmancer docs list
docmancer docs add ./docs
docmancer docs add https://docs.example.com
docmancer docs query "how to authenticate"
docmancer docs query "how to authenticate" --expand
docmancer docs query "how to authenticate" --expand page
docmancer docs query "how to authenticate" --format json
docmancer docs query "how to authenticate" --allow-degraded
docmancer docs sync
docmancer docs remove 
docmancer docs download  --output 
docmancer doctor
```

Use `docs add` for local files, directories, documentation URLs, and GitHub repositories. Use root `import` only for Markdown intended for memory curation.

## Optional providers and scale profile

```bash
docmancer providers list
docmancer providers key 
docmancer providers set  --default --model 
docmancer providers test 

pipx install "docmancer[embeddings-heavy]"
docmancer qdrant up
docmancer setup --profile scale
```

If you use uv, the install line becomes `uv tool install "docmancer[embeddings-heavy]"`. Local retrieval does not require a generation provider or Qdrant.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [docmancer](https://github.com/docmancer)
- **Source:** [docmancer/docmancer](https://github.com/docmancer/docmancer)
- **License:** MIT
- **Homepage:** https://docmancer.dev

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-docmancer-docmancer-docmancer
- Seller: https://agentstack.voostack.com/s/docmancer
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.