ebbtide 0.2.0__tar.gz

ebbtide-0.2.0/.gitignore

@@ -0,0 +1,34 @@
+# Python
+__pycache__/
+*.py[cod]
+*$py.class
+*.so
+*.egg-info/
+build/
+*.egg
+.venv/
+venv/
+env/
+
+# Environment
+.env
+
+# Memory databases (EBBTIDE_DB)
+*.db
+*.db-*
+
+# IDE
+.vscode/
+.idea/
+*.swp
+*.swo
+
+# OS
+.DS_Store
+Thumbs.db
+
+# Local Claude Code settings
+.claude/settings.local.json
+
+# Benchmark datasets and results (bench/): downloaded, never committed
+/data/

ebbtide-0.2.0/LICENSE

@@ -0,0 +1,22 @@
+MIT License
+
+Copyright (c) 2026 Nikhil Pawar
+Copyright (c) 2026 Dory.md contributors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

ebbtide-0.2.0/PKG-INFO

@@ -0,0 +1,151 @@
+Metadata-Version: 2.5
+Name: ebbtide
+Version: 0.2.0
+Summary: Long-term memory for AI agents that forgets on purpose: an MCP server on a forgetting curve.
+Project-URL: Homepage, https://aimsotrash-ebbtide.vercel.app
+Project-URL: Repository, https://github.com/aimsotrash/Ebbtide
+Author: Nikhil Pawar
+License-Expression: MIT
+License-File: LICENSE
+Requires-Python: >=3.12
+Requires-Dist: mcp==2.3.0
+Requires-Dist: numpy==2.4.6
+Requires-Dist: sentence-transformers==6.1.0
+Requires-Dist: torch==2.14.1
+Requires-Dist: transformers==5.19.0
+Description-Content-Type: text/markdown
+
+# Ebbtide
+
+[![CI](https://github.com/aimsotrash/Ebbtide/actions/workflows/ci.yml/badge.svg)](https://github.com/aimsotrash/Ebbtide/actions/workflows/ci.yml)
+[![MCP Badge](https://lobehub.com/badge/mcp/aimsotrash-ebbtide)](https://lobehub.com/mcp/aimsotrash-ebbtide)
+
+**Long-term memory for AI agents that forgets on purpose.**
+
+Ebbtide was called Engram until October 31, 2026; I renamed it because several other memory
+servers for AI agents already use that name. It began as a notes app, which I retired in
+October 2026; its last version is the git tag
+[`notes-app`](https://github.com/aimsotrash/Ebbtide/tree/notes-app).
+
+**Live:** [aimsotrash-ebbtide.vercel.app](https://aimsotrash-ebbtide.vercel.app)
+
+## Install
+
+With [uv](https://docs.astral.sh/uv/), on Linux or an Apple silicon Mac (macOS 14 or later).
+For Claude Code:
+
+```bash
+uv tool install --torch-backend cpu ebbtide
+claude mcp add ebbtide -s user -- ebbtide
+```
+
+For any other MCP client, add this to its configuration:
+
+```json
+{
+  "mcpServers": {
+    "ebbtide": {
+      "command": "uvx",
+      "args": ["--torch-backend", "cpu", "ebbtide"]
+    }
+  }
+}
+```
+
+`--torch-backend cpu` installs PyTorch's CPU build (about 1.2 GB) instead of several GB of
+CUDA libraries; Ebbtide doesn't use a GPU. The first `remember` or `recall` downloads a 90 MB
+embedding model, and after that Ebbtide runs offline. Settings and
+the memory panel are in [`ebbtide/README.md`](ebbtide/README.md).
+
+## Memory for AI agents
+
+[`ebbtide/`](ebbtide/README.md) is an MCP server that gives an agent such as Claude
+Code long-term memory across sessions, through four tools: `remember`, `recall`, `forget`
+and `memory_stats`. Two resources, `ebbtide://memories` and `ebbtide://memories/{id}`, let a
+client attach memories as context without strengthening them, and two prompts,
+`recall_first` and `save_session`, start the common workflows. Each memory fades along a forgetting curve unless it is recalled, and
+every recall makes it last longer. When the store reaches its budget, 1000 memories by
+default, the weakest one goes first. The memories live in a SQLite file on your machine,
+and only what a recall returns enters the agent's context.
+
+A local panel lists the memories and how strong each one is, with a slider that looks
+ahead to show which ones are fading.
+
+### Should an agent forget?
+
+I replayed six memory policies on two public long-conversation benchmarks and measured
+whether the memory still finds the evidence each question needs.
+[LongMemEval](bench/RESULTS.md) has 500 questions with histories of about 490 turns;
+its 30 abstention questions have no evidence to find, so 470 are scored. Settings were picked on
+92 dev questions and fixed before the other 378 were first scored.
+[LoCoMo](bench/RESULTS-locomo.md) has 10 conversations that span 5 to 10 months, and
+nothing was tuned on it.
+
+- **Keeping everything wins when you can afford it.** With room for half the history, the
+  best forgetting policy finds all the evidence in its top 10 results for 37% of LongMemEval
+  questions and 27% of LoCoMo questions. Keeping everything does for 58.5% and 44.6%.
+- **No forgetting rule wins everywhere.** On LongMemEval, keeping what gets used beats
+  keeping the newest. On LoCoMo, keeping the newest does best at the turn level.
+- **Strengthening on recall beats plain LRU, modestly.** ebbtide-count, the server's
+  default, keeps up to 4 points more of the evidence than LRU on LongMemEval, and ranks every
+  answer session in its top 10 up to 14 points more often; I chose these comparisons after
+  seeing the results. On LoCoMo, where I chose the comparison before running it, both gains
+  held but were smaller (up to 1.6 and 3.4 points), and it found all the evidence in its top
+  10 about as often as LRU.
+- **Strengthening too hard backfires on facts that change.** FSRS-style spacing made old
+  facts stick: on LongMemEval's knowledge-update questions it found the newest fact 12 to
+  17 points less often than LRU.
+- **Fading didn't help as a ranking signal.** Tuning picked similarity alone almost
+  everywhere, so the gains come from what is kept, not from how it is ranked.
+
+MemoryBank (Zhong et al., 2023) already applied an Ebbinghaus curve to LLM memory. What is
+new here is the measured comparison with simple baselines, losses included.
+
+## Stack
+
+Python · MCP · sentence-transformers · NumPy · SQLite · React · TypeScript · Vite ·
+Tailwind CSS · Motion
+
+## Run it locally
+
+To run the MCP server, follow the setup in [`ebbtide/README.md`](ebbtide/README.md).
+
+To reproduce the benchmarks, from the repo root:
+
+```bash
+uv venv -p 3.12 .venv && uv pip install --python .venv -e .   # CPU-only PyTorch, per pyproject.toml
+.venv/bin/python -m bench.longmemeval embed   # downloads the data and embeds it once, ~45 min on a laptop CPU
+.venv/bin/python -m bench.longmemeval run     # writes bench/RESULTS.md
+.venv/bin/python -m bench.locomo              # writes bench/RESULTS-locomo.md
+```
+
+The landing page for the MCP server lives in `site/` (Vite, React and Tailwind):
+
+```bash
+cd site && npm install && npm run dev   # http://localhost:5173
+```
+
+The Python tests (20: 13 intelligence, 3 MCP server, 4 benchmark) stub the embedding model,
+so they need only `requirements-test.txt`, which adds pytest; any venv works, such as the one
+above or a fresh `uv venv -p 3.12 .venv`:
+
+```bash
+uv pip install --python .venv -r requirements-test.txt
+.venv/bin/python -m pytest intelligence/tests ebbtide/tests bench
+```
+
+## References
+
+- Di Wu et al. *LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory.*
+  2024. [arXiv:2410.10813](https://arxiv.org/abs/2410.10813)
+- Adyasha Maharana et al. *Evaluating Very Long-Term Conversational Memory of LLM Agents.*
+  ACL 2024. [arXiv:2402.17753](https://arxiv.org/abs/2402.17753). The LoCoMo data is
+  CC BY-NC 4.0; it is downloaded, never committed, and used here for non-commercial research.
+- Wanjun Zhong et al. *MemoryBank: Enhancing Large Language Models with Long-Term Memory.*
+  2023. [arXiv:2305.10250](https://arxiv.org/abs/2305.10250)
+- FSRS, the spaced-repetition scheduler whose default weights the spaced rule uses:
+  [open-spaced-repetition](https://github.com/open-spaced-repetition)
+
+## License
+
+[MIT](LICENSE).

ebbtide-0.2.0/README.md

@@ -0,0 +1,134 @@
+# Ebbtide
+
+[![CI](https://github.com/aimsotrash/Ebbtide/actions/workflows/ci.yml/badge.svg)](https://github.com/aimsotrash/Ebbtide/actions/workflows/ci.yml)
+[![MCP Badge](https://lobehub.com/badge/mcp/aimsotrash-ebbtide)](https://lobehub.com/mcp/aimsotrash-ebbtide)
+
+**Long-term memory for AI agents that forgets on purpose.**
+
+Ebbtide was called Engram until October 31, 2026; I renamed it because several other memory
+servers for AI agents already use that name. It began as a notes app, which I retired in
+October 2026; its last version is the git tag
+[`notes-app`](https://github.com/aimsotrash/Ebbtide/tree/notes-app).
+
+**Live:** [aimsotrash-ebbtide.vercel.app](https://aimsotrash-ebbtide.vercel.app)
+
+## Install
+
+With [uv](https://docs.astral.sh/uv/), on Linux or an Apple silicon Mac (macOS 14 or later).
+For Claude Code:
+
+```bash
+uv tool install --torch-backend cpu ebbtide
+claude mcp add ebbtide -s user -- ebbtide
+```
+
+For any other MCP client, add this to its configuration:
+
+```json
+{
+  "mcpServers": {
+    "ebbtide": {
+      "command": "uvx",
+      "args": ["--torch-backend", "cpu", "ebbtide"]
+    }
+  }
+}
+```
+
+`--torch-backend cpu` installs PyTorch's CPU build (about 1.2 GB) instead of several GB of
+CUDA libraries; Ebbtide doesn't use a GPU. The first `remember` or `recall` downloads a 90 MB
+embedding model, and after that Ebbtide runs offline. Settings and
+the memory panel are in [`ebbtide/README.md`](ebbtide/README.md).
+
+## Memory for AI agents
+
+[`ebbtide/`](ebbtide/README.md) is an MCP server that gives an agent such as Claude
+Code long-term memory across sessions, through four tools: `remember`, `recall`, `forget`
+and `memory_stats`. Two resources, `ebbtide://memories` and `ebbtide://memories/{id}`, let a
+client attach memories as context without strengthening them, and two prompts,
+`recall_first` and `save_session`, start the common workflows. Each memory fades along a forgetting curve unless it is recalled, and
+every recall makes it last longer. When the store reaches its budget, 1000 memories by
+default, the weakest one goes first. The memories live in a SQLite file on your machine,
+and only what a recall returns enters the agent's context.
+
+A local panel lists the memories and how strong each one is, with a slider that looks
+ahead to show which ones are fading.
+
+### Should an agent forget?
+
+I replayed six memory policies on two public long-conversation benchmarks and measured
+whether the memory still finds the evidence each question needs.
+[LongMemEval](bench/RESULTS.md) has 500 questions with histories of about 490 turns;
+its 30 abstention questions have no evidence to find, so 470 are scored. Settings were picked on
+92 dev questions and fixed before the other 378 were first scored.
+[LoCoMo](bench/RESULTS-locomo.md) has 10 conversations that span 5 to 10 months, and
+nothing was tuned on it.
+
+- **Keeping everything wins when you can afford it.** With room for half the history, the
+  best forgetting policy finds all the evidence in its top 10 results for 37% of LongMemEval
+  questions and 27% of LoCoMo questions. Keeping everything does for 58.5% and 44.6%.
+- **No forgetting rule wins everywhere.** On LongMemEval, keeping what gets used beats
+  keeping the newest. On LoCoMo, keeping the newest does best at the turn level.
+- **Strengthening on recall beats plain LRU, modestly.** ebbtide-count, the server's
+  default, keeps up to 4 points more of the evidence than LRU on LongMemEval, and ranks every
+  answer session in its top 10 up to 14 points more often; I chose these comparisons after
+  seeing the results. On LoCoMo, where I chose the comparison before running it, both gains
+  held but were smaller (up to 1.6 and 3.4 points), and it found all the evidence in its top
+  10 about as often as LRU.
+- **Strengthening too hard backfires on facts that change.** FSRS-style spacing made old
+  facts stick: on LongMemEval's knowledge-update questions it found the newest fact 12 to
+  17 points less often than LRU.
+- **Fading didn't help as a ranking signal.** Tuning picked similarity alone almost
+  everywhere, so the gains come from what is kept, not from how it is ranked.
+
+MemoryBank (Zhong et al., 2023) already applied an Ebbinghaus curve to LLM memory. What is
+new here is the measured comparison with simple baselines, losses included.
+
+## Stack
+
+Python · MCP · sentence-transformers · NumPy · SQLite · React · TypeScript · Vite ·
+Tailwind CSS · Motion
+
+## Run it locally
+
+To run the MCP server, follow the setup in [`ebbtide/README.md`](ebbtide/README.md).
+
+To reproduce the benchmarks, from the repo root:
+
+```bash
+uv venv -p 3.12 .venv && uv pip install --python .venv -e .   # CPU-only PyTorch, per pyproject.toml
+.venv/bin/python -m bench.longmemeval embed   # downloads the data and embeds it once, ~45 min on a laptop CPU
+.venv/bin/python -m bench.longmemeval run     # writes bench/RESULTS.md
+.venv/bin/python -m bench.locomo              # writes bench/RESULTS-locomo.md
+```
+
+The landing page for the MCP server lives in `site/` (Vite, React and Tailwind):
+
+```bash
+cd site && npm install && npm run dev   # http://localhost:5173
+```
+
+The Python tests (20: 13 intelligence, 3 MCP server, 4 benchmark) stub the embedding model,
+so they need only `requirements-test.txt`, which adds pytest; any venv works, such as the one
+above or a fresh `uv venv -p 3.12 .venv`:
+
+```bash
+uv pip install --python .venv -r requirements-test.txt
+.venv/bin/python -m pytest intelligence/tests ebbtide/tests bench
+```
+
+## References
+
+- Di Wu et al. *LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory.*
+  2024. [arXiv:2410.10813](https://arxiv.org/abs/2410.10813)
+- Adyasha Maharana et al. *Evaluating Very Long-Term Conversational Memory of LLM Agents.*
+  ACL 2024. [arXiv:2402.17753](https://arxiv.org/abs/2402.17753). The LoCoMo data is
+  CC BY-NC 4.0; it is downloaded, never committed, and used here for non-commercial research.
+- Wanjun Zhong et al. *MemoryBank: Enhancing Large Language Models with Long-Term Memory.*
+  2023. [arXiv:2305.10250](https://arxiv.org/abs/2305.10250)
+- FSRS, the spaced-repetition scheduler whose default weights the spaced rule uses:
+  [open-spaced-repetition](https://github.com/open-spaced-repetition)
+
+## License
+
+[MIT](LICENSE).

ebbtide-0.2.0/ebbtide/README.md

@@ -0,0 +1,111 @@
+# Ebbtide MCP server
+
+Long-term memory for AI agents, served over the Model Context Protocol (stdio). Each
+memory fades along a forgetting curve unless it is recalled, and each recall makes it
+last longer. When the store is full, the weakest memory goes. The memory model and its
+policies live in [`intelligence/memory/agent.py`](../intelligence/memory/agent.py); this
+package adds the tools and SQLite persistence.
+
+## Tools
+
+| Tool | What it does |
+|---|---|
+| `remember(text, tags?)` | Stores one memory. Returns its id and the ids of any memories the budget pushed out. |
+| `recall(query, k=5)` | Returns the `k` most relevant memories (at most 20), each with its strength just before this recall. Recalling strengthens a memory. |
+| `forget(id)` | Deletes a memory permanently. |
+| `memory_stats()` | Reports the count, budget, policy, mean strength, and how many memories have ever been stored. |
+
+## Resources
+
+Reading a resource is not recalling: neither one strengthens a memory.
+
+| Resource | What it holds |
+|---|---|
+| `ebbtide://memories` | Every kept memory, strongest first, with its strength now. |
+| `ebbtide://memories/{id}` | One memory by its id. |
+
+## Prompts
+
+| Prompt | What it asks the agent to do |
+|---|---|
+| `recall_first(question)` | Recall what's relevant before answering the question. |
+| `save_session()` | Store each fact, preference, decision or lesson from the conversation worth keeping. |
+
+## Setup
+
+With [uv](https://docs.astral.sh/uv/), on Linux or an Apple silicon Mac (macOS 14 or later):
+
+```bash
+uv tool install --torch-backend cpu ebbtide
+claude mcp add ebbtide -s user -- ebbtide
+```
+
+The first command installs two commands, `ebbtide` and `ebbtide-panel`, from PyPI in an
+isolated environment. `--torch-backend cpu` picks PyTorch's CPU build (about 1.2 GB on disk)
+over several GB of CUDA libraries; Ebbtide doesn't use a GPU. `-s user` makes Ebbtide available in
+every project; without it, Claude Code would use it only in the folder you ran it from. To
+update later, run `uv tool upgrade ebbtide`. To run the latest commit instead of the latest
+release, install `git+https://github.com/aimsotrash/Ebbtide` in place of `ebbtide`.
+
+Other MCP clients take the same server as a `uvx` command:
+
+```json
+{
+  "mcpServers": {
+    "ebbtide": {
+      "command": "uvx",
+      "args": ["--torch-backend", "cpu", "ebbtide"]
+    }
+  }
+}
+```
+
+`uvx` installs on the first launch, which takes a minute or two and can outlast a client's
+startup timeout. Running `uvx --torch-backend cpu ebbtide < /dev/null` once in a terminal
+does that install ahead of time.
+
+The first `remember` or `recall` in a session takes about 5 seconds while the embedding
+model (`all-MiniLM-L6-v2`) loads, and the very first run also downloads it (~90 MB).
+After that, a call takes about 10 ms.
+
+## Panel
+
+```bash
+ebbtide-panel        # then open http://127.0.0.1:8765
+```
+
+A quiet local page beside the agent: every memory, strongest first, with its strength,
+half-life and recall count. It reads the same database read-only every 2 seconds, so a
+`remember`, a `recall` or an eviction shows up as it happens. The look-ahead slider shows
+how strong each memory would be after that many more days without a recall: the recalled
+ones fade more slowly. It listens on 127.0.0.1 only and loads nothing from the internet.
+
+For a demo that shouldn't touch your real memories, point both at a scratch database: set
+`EBBTIDE_DB=/tmp/ebbtide-demo.db` for the panel, and pass `-e EBBTIDE_DB=/tmp/ebbtide-demo.db`
+when you register the server. Once a first run has downloaded the embedding model, the server
+loads it from disk and never goes online, so venue Wi-Fi doesn't matter.
+
+## Configuration
+
+| Variable | Default | Purpose |
+|---|---|---|
+| `EBBTIDE_DB` | `~/.local/share/ebbtide/memory.db` | SQLite file holding the memories and their embeddings. |
+| `EBBTIDE_POLICY` | `ebbtide-count` | A policy name from `intelligence.memory.POLICIES`. Of the forgetting-curve policies, the default scored best on the benchmark's dev split at the 25% and 50% budgets; see [`bench/RESULTS.md`](../bench/RESULTS.md). |
+| `EBBTIDE_BUDGET` | `1000` | The most memories kept. `keep-all` ignores it. |
+
+Pass them with `-e` when you register the server, and give the panel the same values, since it
+reads them from its own environment. Each tool call is one SQLite
+transaction, so two agents can share one database safely.
+
+## Test
+
+From the repo root:
+
+```bash
+uv venv -p 3.12 .venv && uv pip install --python .venv -r requirements-test.txt
+.venv/bin/python -m pytest ebbtide/tests/
+```
+
+They call every tool through an in-process MCP client and fetch the panel through its
+HTTP handler, with a stub embedder and a fake clock, so they never load the embedding model or need
+PyTorch.

ebbtide-0.2.0/ebbtide/__init__.py

@@ -0,0 +1 @@
+"""Ebbtide as an MCP server: long-term memory for AI agents. Run: python -m ebbtide"""

ebbtide-0.2.0/ebbtide/__main__.py

@@ -0,0 +1,3 @@
+from ebbtide.server import main
+
+main()

ebbtide-0.2.0/ebbtide/fonts/OFL.txt

@@ -0,0 +1,93 @@
+Copyright 2024 The Geist Project Authors (https://github.com/vercel/geist-font) Geist-Italic[wght].ttf: Copyright 2024 The Geist Project Authors (https://github.com/vercel/geist-font)
+
+This Font Software is licensed under the SIL Open Font License, Version 1.1.
+This license is copied below, and is also available with a FAQ at:
+http://scripts.sil.org/OFL
+
+
+-----------------------------------------------------------
+SIL OPEN FONT LICENSE Version 1.1 - 26 February 2007
+-----------------------------------------------------------
+
+PREAMBLE
+The goals of the Open Font License (OFL) are to stimulate worldwide
+development of collaborative font projects, to support the font creation
+efforts of academic and linguistic communities, and to provide a free and
+open framework in which fonts may be shared and improved in partnership
+with others.
+
+The OFL allows the licensed fonts to be used, studied, modified and
+redistributed freely as long as they are not sold by themselves. The
+fonts, including any derivative works, can be bundled, embedded,
+redistributed and/or sold with any software provided that any reserved
+names are not used by derivative works. The fonts and derivatives,
+however, cannot be released under any other type of license. The
+requirement for fonts to remain under this license does not apply
+to any document created using the fonts or their derivatives.
+
+DEFINITIONS
+"Font Software" refers to the set of files released by the Copyright
+Holder(s) under this license and clearly marked as such. This may
+include source files, build scripts and documentation.
+
+"Reserved Font Name" refers to any names specified as such after the
+copyright statement(s).
+
+"Original Version" refers to the collection of Font Software components as
+distributed by the Copyright Holder(s).
+
+"Modified Version" refers to any derivative made by adding to, deleting,
+or substituting -- in part or in whole -- any of the components of the
+Original Version, by changing formats or by porting the Font Software to a
+new environment.
+
+"Author" refers to any designer, engineer, programmer, technical
+writer or other person who contributed to the Font Software.
+
+PERMISSION & CONDITIONS
+Permission is hereby granted, free of charge, to any person obtaining
+a copy of the Font Software, to use, study, copy, merge, embed, modify,
+redistribute, and sell modified and unmodified copies of the Font
+Software, subject to the following conditions:
+
+1) Neither the Font Software nor any of its individual components,
+in Original or Modified Versions, may be sold by itself.
+
+2) Original or Modified Versions of the Font Software may be bundled,
+redistributed and/or sold with any software, provided that each copy
+contains the above copyright notice and this license. These can be
+included either as stand-alone text files, human-readable headers or
+in the appropriate machine-readable metadata fields within text or
+binary files as long as those fields can be easily viewed by the user.
+
+3) No Modified Version of the Font Software may use the Reserved Font
+Name(s) unless explicit written permission is granted by the corresponding
+Copyright Holder. This restriction only applies to the primary font name as
+presented to the users.
+
+4) The name(s) of the Copyright Holder(s) or the Author(s) of the Font
+Software shall not be used to promote, endorse or advertise any
+Modified Version, except to acknowledge the contribution(s) of the
+Copyright Holder(s) and the Author(s) or with their explicit written
+permission.
+
+5) The Font Software, modified or unmodified, in part or in whole,
+must be distributed entirely under this license, and must not be
+distributed under any other license. The requirement for fonts to
+remain under this license does not apply to any document created
+using the Font Software.
+
+TERMINATION
+This license becomes null and void if any of the above conditions are
+not met.
+
+DISCLAIMER
+THE FONT SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTIES OF
+MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT
+OF COPYRIGHT, PATENT, TRADEMARK, OR OTHER RIGHT. IN NO EVENT SHALL THE
+COPYRIGHT HOLDER BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+INCLUDING ANY GENERAL, SPECIAL, INDIRECT, INCIDENTAL, OR CONSEQUENTIAL
+DAMAGES, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+FROM, OUT OF THE USE OR INABILITY TO USE THE FONT SOFTWARE OR FROM
+OTHER DEALINGS IN THE FONT SOFTWARE.