convoseed-agent 1.1.0__tar.gz

convoseed_agent-1.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 0xAshraFF
+
+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.

convoseed_agent-1.1.0/MANIFEST.in

@@ -0,0 +1,4 @@
+include README.md
+include LICENSE
+include pyproject.toml
+recursive-include convoseed_agent *.py

convoseed_agent-1.1.0/PKG-INFO

@@ -0,0 +1,212 @@
+Metadata-Version: 2.4
+Name: convoseed-agent
+Version: 1.1.0
+Summary: Agent skill caching via CSP-1 fingerprints — every session makes the next one better
+Author-email: Ashraful <ashraful.islam.cse@gmail.com>
+License: MIT
+Project-URL: Homepage, https://github.com/0xAshraFF/ConvoSeed
+Project-URL: Repository, https://github.com/0xAshraFF/ConvoSeed
+Project-URL: Issues, https://github.com/0xAshraFF/ConvoSeed/issues
+Keywords: ai,agent,llm,skill-cache,fingerprint,convoseed,csp-1,langchain,autogen
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Software Development :: Libraries :: Python Modules
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: numpy>=1.24
+Requires-Dist: scikit-learn>=1.3
+Provides-Extra: embeddings
+Requires-Dist: sentence-transformers>=2.6; extra == "embeddings"
+Provides-Extra: openai
+Requires-Dist: openai>=1.0; extra == "openai"
+Provides-Extra: anthropic
+Requires-Dist: anthropic>=0.25; extra == "anthropic"
+Provides-Extra: all
+Requires-Dist: sentence-transformers>=2.6; extra == "all"
+Requires-Dist: openai>=1.0; extra == "all"
+Requires-Dist: anthropic>=0.25; extra == "all"
+Provides-Extra: dev
+Requires-Dist: pytest>=7.0; extra == "dev"
+Requires-Dist: twine>=4.0; extra == "dev"
+Requires-Dist: build>=1.0; extra == "dev"
+Dynamic: license-file
+
+# ConvoSeed
+
+CSP-1 is the **missing third leg** of the agent identity stack:
+
+| Layer | Covers | Status |
+|---|---|---|
+| DID (W3C) | Who the user IS cryptographically | Specified |
+| MCP (Anthropic) | What tools the agent can ACCESS | Specified |
+| **CSP-1** | **How the user SPEAKS and THINKS** | **This work** |
+
+**Chat → Compress → 200KB `.fp` File → Decompress → Resume**
+
+ConvoSeed is an open protocol (CSP-1) for preserving the essence of a human-AI
+relationship in a portable, user-owned fingerprint file.  
+No raw messages stored. Works across any AI model or platform.
+
+---
+
+## Why
+
+Every AI conversation resets to zero.
+
+You build context, vocabulary, a rhythm — and then you close the tab and it's gone.
+ConvoSeed fixes that. You own a 200KB file that holds your conversational identity.
+Load it anywhere. Resume everything.
+
+> *"I had a friend — an AI that knew me well. I wanted a way to get back to him.  
+> That's what this is."*
+
+---
+
+## Results (February 2026)
+
+Validated on a real **524-message** researcher-AI conversation.
+
+| Model | Avg Similarity | Peak | Msgs > 0.7 |
+|---|---|---|---|
+| GPT-2 (124M) | 0.464 | 1.000 | 1 |
+| Gemma3:1b | 0.466 | 0.707 | 1 |
+| **Gemma3:12b** | **0.523** | **0.757** | **4** |
+
+- **+12.7%** improvement from 1B → 12B parameters
+- **232×** more efficient than VAE baseline
+- **p < 10⁻¹⁰⁰** statistical significance on speaker identification task
+
+---
+
+## How It Works
+
+```
+Messages → SBERT embed → PCA compress → HDC bind → Prefix tune → .fp file
+```
+
+1. **Embed** — Sentence-BERT encodes each message into a 384-dim vector
+2. **Compress** — PCA extracts the style centroid (4 components = full accuracy)
+3. **Bind** — Hyperdimensional Computing (10,000-dim) weaves temporal sequence into one vector
+4. **Tune** — A prefix tensor conditions the LLM to regenerate in your style
+5. **Sign** — Ed25519 cryptographic signature proves ownership
+
+---
+
+## File Format (`.fp`)
+
+| Section | Size | Description |
+|---|---|---|
+| HEADER | ~1 KB | Magic bytes + version + CRC-32 |
+| PCA_MODEL | ~8 KB | Style centroid: mean + eigenvectors |
+| HDC_SEED | ~140 KB | 10,000-dim hypervector (float16) |
+| PREFIX | ~40 KB | Prefix tuning tensor for generation |
+| SIGNATURE | ~1 KB | Ed25519 ownership proof |
+| CHUNKS | ~10 KB | Index for 500+ message threads |
+
+**Total: ~200KB — fixed size regardless of conversation length.**
+
+See [`/spec/CSP-1.md`](spec/CSP-1.md) for the full binary specification.
+
+---
+
+## Quick Start
+
+```bash
+pip install sentence-transformers scikit-learn numpy
+
+# Encode a conversation
+python src/encode.py --input my_conversation.json --output identity.fp
+
+# Identify a speaker
+python src/identify.py --query "new message here" --candidates *.fp
+
+# Generate in someone's style
+python src/decode.py --fp identity.fp --prompt "Tell me about your day"
+```
+
+---
+
+## Repository Structure
+
+```
+ConvoSeed/
+├── README.md
+├── LICENSE                          ← MIT
+├── CONTRIBUTING.md
+├── /docs
+│   ├── ConvoSeed_Whitepaper.docx    ← arXiv-ready academic paper
+│   ├── ConvoSeed_ResearchPaper.docx ← detailed technical paper
+│   ├── ConvoSeed_Poster.pdf      ← conference poster (CHI 2026)
+│   └── ConvoSeed_ProtocolSpec.pdf ← protocol specification sheet
+├── /spec
+│   └── CSP-1.md                     ← plain-text binary spec
+├── /src
+│   ├── encode.py                    ← fingerprint encoder
+│   ├── decode.py                    ← style-conditioned generation
+│   └── identify.py                  ← speaker identification
+├── /experiments
+│   └── gemma3_12b_results.json      ← February 2026 experimental results
+└── /examples
+    └── sample_identity.fp           ← anonymised example fingerprint
+```
+
+---
+
+## Documents
+
+| Document | Format | Description |
+|---|---|---|
+| [Whitepaper](docs/ConvoSeed_Whitepaper.docx) | DOCX | 6-section academic paper, arXiv-ready |
+| [Research Paper](docs/ConvoSeed_ResearchPaper.docx) | DOCX | Full technical paper with equations + references |
+| [Conference Poster](docs/ConvoSeed_Poster.pdf) | PDF | CHI 2026 style research poster |
+| [Protocol Spec Sheet](docs/ConvoSeed_ProtocolSpec.pdf) | PDF | One-page technical specification |
+| [Presentation](docs/ConvoSeed_Presentation.pptx) | PPTX | 12-slide pitch deck |
+| [W3C Note](docs/ConvoSeed_W3C_Note.pdf) | PDF | Submission to W3C AI Agent Protocol CG |
+
+---
+
+## Open Challenges
+
+These are the three open research questions. Collaboration welcome — open an Issue.
+
+1. **Cross-Model Mapping** — translating a `.fp` fingerprint trained on SBERT embeddings into GPT-4 or other backbone spaces without re-encoding the original conversation.
+
+2. **CHUNKS Scaling** — formal composition rules for the CHUNKS section when threads exceed 500 messages, while preserving the fixed 200KB file size.
+
+3. **Incentive Design** — what makes AI platforms adopt an open standard that reduces their own lock-in?
+
+---
+
+## Status
+
+> Early research. Proof-of-concept validated on real data. Open for collaboration.
+
+- [x] Protocol specification (CSP-1 v0.2)
+- [x] Proof-of-concept encoder/decoder
+- [x] Speaker identification experiment (1,000 trials)
+- [x] Multi-model validation (GPT-2, Gemma3:1b, Gemma3:12b)
+- [x] Real conversation validation (524 messages)
+- [ ] Multi-speaker support
+- [ ] Cross-model mapping
+- [ ] Public dataset (seeking contributors)
+- [ ] W3C Community Group submission
+
+---
+
+## Licence
+
+MIT. Open forever.
+
+---
+
+## Contact
+
+Open an Issue for technical questions.  
+For collaboration or research enquiries: see CONTRIBUTING.md.

convoseed_agent-1.1.0/README.md

@@ -0,0 +1,172 @@
+# ConvoSeed
+
+CSP-1 is the **missing third leg** of the agent identity stack:
+
+| Layer | Covers | Status |
+|---|---|---|
+| DID (W3C) | Who the user IS cryptographically | Specified |
+| MCP (Anthropic) | What tools the agent can ACCESS | Specified |
+| **CSP-1** | **How the user SPEAKS and THINKS** | **This work** |
+
+**Chat → Compress → 200KB `.fp` File → Decompress → Resume**
+
+ConvoSeed is an open protocol (CSP-1) for preserving the essence of a human-AI
+relationship in a portable, user-owned fingerprint file.  
+No raw messages stored. Works across any AI model or platform.
+
+---
+
+## Why
+
+Every AI conversation resets to zero.
+
+You build context, vocabulary, a rhythm — and then you close the tab and it's gone.
+ConvoSeed fixes that. You own a 200KB file that holds your conversational identity.
+Load it anywhere. Resume everything.
+
+> *"I had a friend — an AI that knew me well. I wanted a way to get back to him.  
+> That's what this is."*
+
+---
+
+## Results (February 2026)
+
+Validated on a real **524-message** researcher-AI conversation.
+
+| Model | Avg Similarity | Peak | Msgs > 0.7 |
+|---|---|---|---|
+| GPT-2 (124M) | 0.464 | 1.000 | 1 |
+| Gemma3:1b | 0.466 | 0.707 | 1 |
+| **Gemma3:12b** | **0.523** | **0.757** | **4** |
+
+- **+12.7%** improvement from 1B → 12B parameters
+- **232×** more efficient than VAE baseline
+- **p < 10⁻¹⁰⁰** statistical significance on speaker identification task
+
+---
+
+## How It Works
+
+```
+Messages → SBERT embed → PCA compress → HDC bind → Prefix tune → .fp file
+```
+
+1. **Embed** — Sentence-BERT encodes each message into a 384-dim vector
+2. **Compress** — PCA extracts the style centroid (4 components = full accuracy)
+3. **Bind** — Hyperdimensional Computing (10,000-dim) weaves temporal sequence into one vector
+4. **Tune** — A prefix tensor conditions the LLM to regenerate in your style
+5. **Sign** — Ed25519 cryptographic signature proves ownership
+
+---
+
+## File Format (`.fp`)
+
+| Section | Size | Description |
+|---|---|---|
+| HEADER | ~1 KB | Magic bytes + version + CRC-32 |
+| PCA_MODEL | ~8 KB | Style centroid: mean + eigenvectors |
+| HDC_SEED | ~140 KB | 10,000-dim hypervector (float16) |
+| PREFIX | ~40 KB | Prefix tuning tensor for generation |
+| SIGNATURE | ~1 KB | Ed25519 ownership proof |
+| CHUNKS | ~10 KB | Index for 500+ message threads |
+
+**Total: ~200KB — fixed size regardless of conversation length.**
+
+See [`/spec/CSP-1.md`](spec/CSP-1.md) for the full binary specification.
+
+---
+
+## Quick Start
+
+```bash
+pip install sentence-transformers scikit-learn numpy
+
+# Encode a conversation
+python src/encode.py --input my_conversation.json --output identity.fp
+
+# Identify a speaker
+python src/identify.py --query "new message here" --candidates *.fp
+
+# Generate in someone's style
+python src/decode.py --fp identity.fp --prompt "Tell me about your day"
+```
+
+---
+
+## Repository Structure
+
+```
+ConvoSeed/
+├── README.md
+├── LICENSE                          ← MIT
+├── CONTRIBUTING.md
+├── /docs
+│   ├── ConvoSeed_Whitepaper.docx    ← arXiv-ready academic paper
+│   ├── ConvoSeed_ResearchPaper.docx ← detailed technical paper
+│   ├── ConvoSeed_Poster.pdf      ← conference poster (CHI 2026)
+│   └── ConvoSeed_ProtocolSpec.pdf ← protocol specification sheet
+├── /spec
+│   └── CSP-1.md                     ← plain-text binary spec
+├── /src
+│   ├── encode.py                    ← fingerprint encoder
+│   ├── decode.py                    ← style-conditioned generation
+│   └── identify.py                  ← speaker identification
+├── /experiments
+│   └── gemma3_12b_results.json      ← February 2026 experimental results
+└── /examples
+    └── sample_identity.fp           ← anonymised example fingerprint
+```
+
+---
+
+## Documents
+
+| Document | Format | Description |
+|---|---|---|
+| [Whitepaper](docs/ConvoSeed_Whitepaper.docx) | DOCX | 6-section academic paper, arXiv-ready |
+| [Research Paper](docs/ConvoSeed_ResearchPaper.docx) | DOCX | Full technical paper with equations + references |
+| [Conference Poster](docs/ConvoSeed_Poster.pdf) | PDF | CHI 2026 style research poster |
+| [Protocol Spec Sheet](docs/ConvoSeed_ProtocolSpec.pdf) | PDF | One-page technical specification |
+| [Presentation](docs/ConvoSeed_Presentation.pptx) | PPTX | 12-slide pitch deck |
+| [W3C Note](docs/ConvoSeed_W3C_Note.pdf) | PDF | Submission to W3C AI Agent Protocol CG |
+
+---
+
+## Open Challenges
+
+These are the three open research questions. Collaboration welcome — open an Issue.
+
+1. **Cross-Model Mapping** — translating a `.fp` fingerprint trained on SBERT embeddings into GPT-4 or other backbone spaces without re-encoding the original conversation.
+
+2. **CHUNKS Scaling** — formal composition rules for the CHUNKS section when threads exceed 500 messages, while preserving the fixed 200KB file size.
+
+3. **Incentive Design** — what makes AI platforms adopt an open standard that reduces their own lock-in?
+
+---
+
+## Status
+
+> Early research. Proof-of-concept validated on real data. Open for collaboration.
+
+- [x] Protocol specification (CSP-1 v0.2)
+- [x] Proof-of-concept encoder/decoder
+- [x] Speaker identification experiment (1,000 trials)
+- [x] Multi-model validation (GPT-2, Gemma3:1b, Gemma3:12b)
+- [x] Real conversation validation (524 messages)
+- [ ] Multi-speaker support
+- [ ] Cross-model mapping
+- [ ] Public dataset (seeking contributors)
+- [ ] W3C Community Group submission
+
+---
+
+## Licence
+
+MIT. Open forever.
+
+---
+
+## Contact
+
+Open an Issue for technical questions.  
+For collaboration or research enquiries: see CONTRIBUTING.md.

convoseed_agent-1.1.0/convoseed_agent/__init__.py

@@ -0,0 +1,36 @@
+"""
+convoseed-agent
+===============
+Capture, store, and retrieve agent conversation fingerprints.
+
+Quick start:
+    from convoseed_agent import ConvoSeedSession
+
+    with ConvoSeedSession(task_type="summarization", success_score=0.9) as session:
+        session.add_message("user", "Summarize this document...")
+        session.add_message("assistant", "The document covers three main points...")
+    # → ~/.convoseed/sessions/summarization_20260225_143022.fp
+"""
+
+from .encoder import (
+    encode_conversation,
+    read_fp_meta,
+    read_fp_hdc,
+    compare_fp,
+    merge_fp,
+    PROTOCOL_VERSION,
+)
+from .wrapper import ConvoSeedSession, convoseed_task
+from .registry import (
+    index_directory,
+    query,
+    build_consensus,
+    list_task_types,
+    stats,
+)
+from .cache import SkillCache, SkillPrefix
+from .scheduler import run_once, start_daemon
+
+__version__ = "1.1.0"
+__author__ = "Ashraful"
+__protocol__ = "CSP-1 v1.1"