recall-echo 4.7.0

Persistent memory system with knowledge graph — for any LLM tool
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65

<p align="center">
  <img src="docs/banner.svg" width="720" alt="recall-echo: persistent memory for AI coding agents">
</p>

<p align="center">
  <a href="https://crates.io/crates/recall-echo"><img src="https://img.shields.io/crates/v/recall-echo?color=2ea44f" alt="crates.io"></a>
  <a href="https://github.com/dnacenta/recall-echo/actions/workflows/ci.yml"><img src="https://github.com/dnacenta/recall-echo/actions/workflows/ci.yml/badge.svg" alt="ci"></a>
  <img src="https://img.shields.io/badge/platforms-macOS%20%7C%20Linux-informational" alt="platforms">
  <a href="LICENSE"><img src="https://img.shields.io/badge/license-MPL--2.0-blue" alt="license"></a>
</p>

<p align="center">
  Memory for your coding agent: every session kept, turned into a knowledge graph,
  and there to ask the next time it matters.
</p>

recall-echo captures your agent's sessions as they end, turns them into a knowledge graph
while the machine is idle, and gives every claim a confidence that rises with corroboration
and falls with contradiction. Your agent asks it questions over MCP. The database and the
embeddings run locally; only extraction calls a model, and that can be the CLI subscription
you already pay for.

## Install

```sh
curl -fsSL https://raw.githubusercontent.com/dnacenta/recall-echo/main/install.sh | sh
recall-echo init
```

`init` finds your agent CLIs (Claude Code, Codex, Grok, Gemini), installs the hooks,
registers the MCP server with each and downloads the embedding model. With cargo:
`cargo install recall-echo --locked`. Prebuilt binaries are on the
[releases page](https://github.com/dnacenta/recall-echo/releases/latest).

## Using it

Day to day you do nothing. To look, or to fix something:

```sh
recall-echo what-do-you-know --about D          # what it believes, and how sure it is
recall-echo graph query "auth flow"             # search by meaning
recall-echo graph correct D USES Vim --wrong    # that claim is mistaken
recall-echo status                              # health
```

Inside the agent, it asks for itself:

> **You:** why did we drop the websocket approach?
> **Agent:** *(calls `recall_query`)* You moved off it in June: the reconnect logic kept
> dropping messages under load, and you settled on polling every 30 seconds.

On a LongMemEval subset it went from 0 of 9 answers right to 7 of 9. A small sample,
stated as one: [benchmarks](docs/benchmarks/).

## Docs

- [How it works](docs/internals.md): the four layers, confidence, provenance, MCP, the daemon
- [Commands](docs/commands.md): every subcommand and flag
- [Configuration](docs/configuration.md): providers and every setting
- [Troubleshooting](docs/troubleshooting.md)

## License

[MPL-2.0](LICENSE). Versions up to v3.13.0 remain available under AGPL-3.0. The graph
store, SurrealDB 3.x, is BSL 1.1: embedding it is allowed, and it becomes Apache-2.0 in 2030.