cognee 1.2.2.dev0__tar.gz → 1.2.2.dev2__tar.gz

{cognee-1.2.2.dev0 → cognee-1.2.2.dev2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cognee
-Version: 1.2.2.dev0
+Version: 1.2.2.dev2
 Summary: Cognee - is a library for enriching LLM context with a semantic layer for better understanding and reasoning.
 Project-URL: Homepage, https://www.cognee.ai
 Project-URL: Repository, https://github.com/topoteretes/cognee
@@ -226,10 +226,9 @@ Cognee is the open-source AI memory platform that gives AI agents persistent lon
   <a href="https://www.readme-i18n.com/topoteretes/cognee?lang=zh">中文</a>
   </p>
 
-
-<div style="text-align: center">
-  <img src="https://raw.githubusercontent.com/topoteretes/cognee/refs/heads/main/assets/cognee_benefits.png" alt="Why cognee?" width="80%" />
-</div>
+<p align="center">
+  <img src="assets/cognee-demo.gif" alt="Cognee Demo" width="80%" />
+</p>
 </div>
 
 📄 Read the research paper: [Optimizing the Interface Between Knowledge Graphs and LLMs for Complex Reasoning](https://arxiv.org/abs/2505.24478) — Markovic et al., 2025
@@ -247,6 +246,10 @@ Cognee is an open-source AI memory platform for AI Agents. Ingest data in any fo
 
 ✴️ _Available as a plugin for your Claude Code — [claude-code-plugin](https://github.com/topoteretes/cognee-integrations/tree/main/integrations/claude-code)_
 
+🦀 _Available as a Rust client — [cognee-rs](https://github.com/topoteretes/cognee-rs)_
+
+🟦 _Available as a TypeScript client — [@cognee/cognee-ts](https://www.npmjs.com/package/@cognee/cognee-ts)_
+
 
 
 ### Why use Cognee:
@@ -256,10 +259,14 @@ Cognee is an open-source AI memory platform for AI Agents. Ingest data in any fo
 - Persistent and Learning Agents - learn from feedback, context management, cross-agent knowledge sharing
 - Reliable and Trustworthy Agents - agentic user/tenant isolation, traceability, OTEL collector, audit traits
 
-### Product Features
+### How it Works
 
 <p align="center">
-  <img src="assets/cognee_products.png" alt="Cognee Products" width="80%" />
+  <img src="assets/remember.svg" alt="Cognee Products" width="80%" />
+</p>
+
+<p align="center">
+  <img src="assets/recall.svg" alt="Cognee Recall" width="80%" />
 </p>
 
 ## Basic Usage & Feature Guide
@@ -394,32 +401,38 @@ extras, and MCP client configuration.
 
 ### Claude Code
 
-Install the [Cognee memory plugin](https://github.com/topoteretes/cognee-integrations/tree/main/integrations/claude-code) to give Claude Code persistent memory across sessions. The plugin automatically captures tool calls into session memory via hooks and syncs to the permanent knowledge graph at session end.
+Install the [Cognee memory plugin](https://github.com/topoteretes/cognee-integrations/tree/main/integrations/claude-code) to give Claude Code persistent memory across sessions. The plugin captures prompts, tool traces, and assistant responses into session memory, injects relevant context on every prompt, and syncs session memory into the permanent knowledge graph at session end.
 
-**Setup:**
+**Install** from the Claude Code marketplace. The recommended way is from your shell, *before* launching Claude Code, so the first `claude` launch is a clean session that bootstraps memory automatically:
 
 ```bash
-# Install cognee
-pip install cognee
+# Add the marketplace and install the plugin (one-time, user-scoped)
+claude plugin marketplace add topoteretes/cognee-integrations
+claude plugin install cognee-memory@cognee
 
-# Configure
-export LLM_API_KEY="your-openai-key"
+# Set env vars for your mode (see below), then launch
+export LLM_API_KEY="sk-..."   # local mode; or COGNEE_BASE_URL + COGNEE_API_KEY for cloud
+claude
+```
 
-# Clone the plugin
-git clone https://github.com/topoteretes/cognee-integrations.git
+**Local mode** (default) — the plugin bootstraps a local Cognee API at `http://localhost:8011`. Only `LLM_API_KEY` is required; the Cognee API key is auto-minted if absent:
 
-# Enable it (add to ~/.zshrc for permanent use)
-claude --plugin-dir ./cognee-integrations/integrations/claude-code
+```bash
+export LLM_API_KEY="sk-..."
 ```
 
-Or connect to Cognee Cloud instead of running locally:
+**Cognee Cloud or a remote server** — set both:
 
 ```bash
-export COGNEE_SERVICE_URL="https://your-instance.cognee.ai"
+export COGNEE_BASE_URL="https://your-instance.cognee.ai"
 export COGNEE_API_KEY="ck_..."
 ```
 
-The plugin hooks into Claude Code's lifecycle — `SessionStart` initializes memory, `PostToolUse` captures actions, `UserPromptSubmit` injects relevant context, `PreCompact` preserves memory across context resets, and `SessionEnd` bridges session data into the permanent graph.
+On startup you should see a "Cognee Memory Connected" system message.
+
+The plugin hooks into Claude Code's lifecycle — `SessionStart` selects mode and sets up identity, `UserPromptSubmit` injects dataset-scoped context, `PostToolUse` captures tool traces, `Stop` writes the assistant's answer, `PreCompact` preserves memory across context resets, and `SessionEnd` triggers the final sync into the permanent graph.
+
+See the [plugin README](https://github.com/topoteretes/cognee-integrations/tree/main/integrations/claude-code) for sessions, datasets, and full configuration.
 
 ### Connect to Cognee Cloud
 
@@ -484,6 +497,38 @@ Agent: "Here's how senior analysts solved a similar retention query.
 - Updates memory with new successful patterns so junior analysts perform at near-expert level
 ```
 
+## Run the Whole Memory Layer on Postgres
+
+Graph memory traditionally means operating a stack — a graph database for relationships, a vector database for embeddings, Redis for sessions, and a relational database for metadata — all deployed, secured, and paid for before an agent remembers anything. In cognee 1.0 you can run the entire memory layer on a single Postgres instance.
+
+| Memory layer | Traditional stack | cognee on Postgres |
+| --- | --- | --- |
+| Relationships | Neo4j or another graph database | cognee's Postgres graph backend |
+| Embeddings | Dedicated vector database | pgvector |
+| Sessions | Redis | SQL session-cache backend |
+| Metadata | Relational database | same Postgres |
+
+The graph still exists — it just lives inside the same Postgres-backed memory layer as the text, metadata, and embeddings, so retrieval moves between similarity and structure without crossing service boundaries. In our CI benchmarks, Postgres search ran ~10% faster than the separate graph-plus-vector setup.
+
+Postgres is the default we recommend for most deployments, but you can still swap in dedicated backends when a workload needs them (Neo4j and Neptune for graphs, Redis for sessions, pgvector and LanceDB for vectors, plus Qdrant, ChromaDB, Weaviate, and Milvus via community adapters). Local development stays fully embedded — SQLite, LanceDB, and Kuzudb — with no extra services to stand up.
+
+```bash
+pip install "cognee[postgres]"
+```
+
+```bash
+DB_PROVIDER=postgres
+VECTOR_DB_PROVIDER=pgvector
+GRAPH_DATABASE_PROVIDER=postgres
+CACHE_BACKEND=postgres
+
+DB_HOST=localhost
+DB_PORT=5432
+DB_USERNAME=cognee
+DB_PASSWORD=cognee
+DB_NAME=cognee_db
+```
+
 ## Deploy Cognee
 
 Use [Cognee Cloud](https://www.cognee.ai) for a fully managed experience, or self-host with one of the 1-click deployment configurations below.
@@ -499,6 +544,41 @@ Use [Cognee Cloud](https://www.cognee.ai) for a fully managed experience, or sel
 
 See the [`distributed/`](distributed/) folder for deploy scripts, worker configurations, and additional details.
 
+## Use Cognee in Other Languages
+
+Prefer something other than Python? Cognee also ships official clients for Rust and TypeScript.
+
+### Getting Started with Rust
+
+Use the [cognee-rs](https://github.com/topoteretes/cognee-rs) crate to add, cognify, and search from Rust.
+
+```bash
+cargo add cognee
+```
+
+See the [cognee-rs repository](https://github.com/topoteretes/cognee-rs) for full setup and examples.
+
+### Getting Started with TypeScript
+
+Use the [@cognee/cognee-ts](https://www.npmjs.com/package/@cognee/cognee-ts) package to add, cognify, and search from Node.js or the browser.
+
+```bash
+npm install @cognee/cognee-ts
+```
+
+See the [@cognee/cognee-ts package](https://www.npmjs.com/package/@cognee/cognee-ts) for full setup and examples.
+
+## Benchmarks
+
+We ran cognee against [BEAM](https://github.com/topoteretes/cognee), a long-context benchmark that tests whether a system can keep track of a long conversation as it changes — a more useful test for agent memory than typical needle-in-a-haystack benchmarks. Using only cognee's default settings and standard open-source features (no custom models, no BEAM-specific pipelines), we beat the previous state of the art at the 100K-token setting and matched it at 10M tokens.
+
+| Benchmark | Setting | cognee | Previous SOTA | Obsidian / RAG baseline |
+|-----------|---------|--------|---------------|--------------------------|
+| BEAM | 100K tokens | **0.79** (>0.8 with per-question routing) | 0.735 | ~0.33 |
+| BEAM | 10M tokens | **0.67** | 0.641 | ~0.33 |
+
+These numbers are a directional signal rather than a definitive measure — see the write-up for the full methodology, caveats, and what the results actually mean.
+
 ## Latest News
 
 [![Watch Demo](https://img.youtube.com/vi/8hmqS2Y5RVQ/maxresdefault.jpg)](https://www.youtube.com/watch?v=8hmqS2Y5RVQ&t=13s)

{cognee-1.2.2.dev0 → cognee-1.2.2.dev2}/README.md

@@ -53,10 +53,9 @@ Cognee is the open-source AI memory platform that gives AI agents persistent lon
   <a href="https://www.readme-i18n.com/topoteretes/cognee?lang=zh">中文</a>
   </p>
 
-
-<div style="text-align: center">
-  <img src="https://raw.githubusercontent.com/topoteretes/cognee/refs/heads/main/assets/cognee_benefits.png" alt="Why cognee?" width="80%" />
-</div>
+<p align="center">
+  <img src="assets/cognee-demo.gif" alt="Cognee Demo" width="80%" />
+</p>
 </div>
 
 📄 Read the research paper: [Optimizing the Interface Between Knowledge Graphs and LLMs for Complex Reasoning](https://arxiv.org/abs/2505.24478) — Markovic et al., 2025
@@ -74,6 +73,10 @@ Cognee is an open-source AI memory platform for AI Agents. Ingest data in any fo
 
 ✴️ _Available as a plugin for your Claude Code — [claude-code-plugin](https://github.com/topoteretes/cognee-integrations/tree/main/integrations/claude-code)_
 
+🦀 _Available as a Rust client — [cognee-rs](https://github.com/topoteretes/cognee-rs)_
+
+🟦 _Available as a TypeScript client — [@cognee/cognee-ts](https://www.npmjs.com/package/@cognee/cognee-ts)_
+
 
 
 ### Why use Cognee:
@@ -83,10 +86,14 @@ Cognee is an open-source AI memory platform for AI Agents. Ingest data in any fo
 - Persistent and Learning Agents - learn from feedback, context management, cross-agent knowledge sharing
 - Reliable and Trustworthy Agents - agentic user/tenant isolation, traceability, OTEL collector, audit traits
 
-### Product Features
+### How it Works
 
 <p align="center">
-  <img src="assets/cognee_products.png" alt="Cognee Products" width="80%" />
+  <img src="assets/remember.svg" alt="Cognee Products" width="80%" />
+</p>
+
+<p align="center">
+  <img src="assets/recall.svg" alt="Cognee Recall" width="80%" />
 </p>
 
 ## Basic Usage & Feature Guide
@@ -221,32 +228,38 @@ extras, and MCP client configuration.
 
 ### Claude Code
 
-Install the [Cognee memory plugin](https://github.com/topoteretes/cognee-integrations/tree/main/integrations/claude-code) to give Claude Code persistent memory across sessions. The plugin automatically captures tool calls into session memory via hooks and syncs to the permanent knowledge graph at session end.
+Install the [Cognee memory plugin](https://github.com/topoteretes/cognee-integrations/tree/main/integrations/claude-code) to give Claude Code persistent memory across sessions. The plugin captures prompts, tool traces, and assistant responses into session memory, injects relevant context on every prompt, and syncs session memory into the permanent knowledge graph at session end.
 
-**Setup:**
+**Install** from the Claude Code marketplace. The recommended way is from your shell, *before* launching Claude Code, so the first `claude` launch is a clean session that bootstraps memory automatically:
 
 ```bash
-# Install cognee
-pip install cognee
+# Add the marketplace and install the plugin (one-time, user-scoped)
+claude plugin marketplace add topoteretes/cognee-integrations
+claude plugin install cognee-memory@cognee
 
-# Configure
-export LLM_API_KEY="your-openai-key"
+# Set env vars for your mode (see below), then launch
+export LLM_API_KEY="sk-..."   # local mode; or COGNEE_BASE_URL + COGNEE_API_KEY for cloud
+claude
+```
 
-# Clone the plugin
-git clone https://github.com/topoteretes/cognee-integrations.git
+**Local mode** (default) — the plugin bootstraps a local Cognee API at `http://localhost:8011`. Only `LLM_API_KEY` is required; the Cognee API key is auto-minted if absent:
 
-# Enable it (add to ~/.zshrc for permanent use)
-claude --plugin-dir ./cognee-integrations/integrations/claude-code
+```bash
+export LLM_API_KEY="sk-..."
 ```
 
-Or connect to Cognee Cloud instead of running locally:
+**Cognee Cloud or a remote server** — set both:
 
 ```bash
-export COGNEE_SERVICE_URL="https://your-instance.cognee.ai"
+export COGNEE_BASE_URL="https://your-instance.cognee.ai"
 export COGNEE_API_KEY="ck_..."
 ```
 
-The plugin hooks into Claude Code's lifecycle — `SessionStart` initializes memory, `PostToolUse` captures actions, `UserPromptSubmit` injects relevant context, `PreCompact` preserves memory across context resets, and `SessionEnd` bridges session data into the permanent graph.
+On startup you should see a "Cognee Memory Connected" system message.
+
+The plugin hooks into Claude Code's lifecycle — `SessionStart` selects mode and sets up identity, `UserPromptSubmit` injects dataset-scoped context, `PostToolUse` captures tool traces, `Stop` writes the assistant's answer, `PreCompact` preserves memory across context resets, and `SessionEnd` triggers the final sync into the permanent graph.
+
+See the [plugin README](https://github.com/topoteretes/cognee-integrations/tree/main/integrations/claude-code) for sessions, datasets, and full configuration.
 
 ### Connect to Cognee Cloud
 
@@ -311,6 +324,38 @@ Agent: "Here's how senior analysts solved a similar retention query.
 - Updates memory with new successful patterns so junior analysts perform at near-expert level
 ```
 
+## Run the Whole Memory Layer on Postgres
+
+Graph memory traditionally means operating a stack — a graph database for relationships, a vector database for embeddings, Redis for sessions, and a relational database for metadata — all deployed, secured, and paid for before an agent remembers anything. In cognee 1.0 you can run the entire memory layer on a single Postgres instance.
+
+| Memory layer | Traditional stack | cognee on Postgres |
+| --- | --- | --- |
+| Relationships | Neo4j or another graph database | cognee's Postgres graph backend |
+| Embeddings | Dedicated vector database | pgvector |
+| Sessions | Redis | SQL session-cache backend |
+| Metadata | Relational database | same Postgres |
+
+The graph still exists — it just lives inside the same Postgres-backed memory layer as the text, metadata, and embeddings, so retrieval moves between similarity and structure without crossing service boundaries. In our CI benchmarks, Postgres search ran ~10% faster than the separate graph-plus-vector setup.
+
+Postgres is the default we recommend for most deployments, but you can still swap in dedicated backends when a workload needs them (Neo4j and Neptune for graphs, Redis for sessions, pgvector and LanceDB for vectors, plus Qdrant, ChromaDB, Weaviate, and Milvus via community adapters). Local development stays fully embedded — SQLite, LanceDB, and Kuzudb — with no extra services to stand up.
+
+```bash
+pip install "cognee[postgres]"
+```
+
+```bash
+DB_PROVIDER=postgres
+VECTOR_DB_PROVIDER=pgvector
+GRAPH_DATABASE_PROVIDER=postgres
+CACHE_BACKEND=postgres
+
+DB_HOST=localhost
+DB_PORT=5432
+DB_USERNAME=cognee
+DB_PASSWORD=cognee
+DB_NAME=cognee_db
+```
+
 ## Deploy Cognee
 
 Use [Cognee Cloud](https://www.cognee.ai) for a fully managed experience, or self-host with one of the 1-click deployment configurations below.
@@ -326,6 +371,41 @@ Use [Cognee Cloud](https://www.cognee.ai) for a fully managed experience, or sel
 
 See the [`distributed/`](distributed/) folder for deploy scripts, worker configurations, and additional details.
 
+## Use Cognee in Other Languages
+
+Prefer something other than Python? Cognee also ships official clients for Rust and TypeScript.
+
+### Getting Started with Rust
+
+Use the [cognee-rs](https://github.com/topoteretes/cognee-rs) crate to add, cognify, and search from Rust.
+
+```bash
+cargo add cognee
+```
+
+See the [cognee-rs repository](https://github.com/topoteretes/cognee-rs) for full setup and examples.
+
+### Getting Started with TypeScript
+
+Use the [@cognee/cognee-ts](https://www.npmjs.com/package/@cognee/cognee-ts) package to add, cognify, and search from Node.js or the browser.
+
+```bash
+npm install @cognee/cognee-ts
+```
+
+See the [@cognee/cognee-ts package](https://www.npmjs.com/package/@cognee/cognee-ts) for full setup and examples.
+
+## Benchmarks
+
+We ran cognee against [BEAM](https://github.com/topoteretes/cognee), a long-context benchmark that tests whether a system can keep track of a long conversation as it changes — a more useful test for agent memory than typical needle-in-a-haystack benchmarks. Using only cognee's default settings and standard open-source features (no custom models, no BEAM-specific pipelines), we beat the previous state of the art at the 100K-token setting and matched it at 10M tokens.
+
+| Benchmark | Setting | cognee | Previous SOTA | Obsidian / RAG baseline |
+|-----------|---------|--------|---------------|--------------------------|
+| BEAM | 100K tokens | **0.79** (>0.8 with per-question routing) | 0.735 | ~0.33 |
+| BEAM | 10M tokens | **0.67** | 0.641 | ~0.33 |
+
+These numbers are a directional signal rather than a definitive measure — see the write-up for the full methodology, caveats, and what the results actually mean.
+
 ## Latest News
 
 [![Watch Demo](https://img.youtube.com/vi/8hmqS2Y5RVQ/maxresdefault.jpg)](https://www.youtube.com/watch?v=8hmqS2Y5RVQ&t=13s)

cognee-1.2.2.dev2/assets/recall.svg

@@ -0,0 +1,70 @@
+<svg viewBox="0 0 740 400" xmlns="http://www.w3.org/2000/svg" role="img" font-family="ui-sans-serif, system-ui, -apple-system, Segoe UI, Roboto, Arial, sans-serif">
+<title>recall() answering a question from the memory graph</title>
+<desc>cognee.recall() reads "Where is cognee based?", lights up the path from the cognee node along the "based in" edge to Berlin, and returns the grounded answer "cognee is based in Berlin".</desc>
+<defs>
+<filter id="glow" x="-80%" y="-80%" width="260%" height="260%"><feGaussianBlur stdDeviation="4" result="b"/><feMerge><feMergeNode in="b"/><feMergeNode in="SourceGraphic"/></feMerge></filter>
+<filter id="softglow" x="-120%" y="-120%" width="340%" height="340%"><feGaussianBlur stdDeviation="7" result="b"/><feMerge><feMergeNode in="b"/><feMergeNode in="SourceGraphic"/></feMerge></filter>
+</defs>
+
+<rect x="0" y="0" width="740" height="400" fill="#0a0a0d"/>
+
+<text x="40" y="42" font-size="21" font-weight="700" fill="#f3f0fa">Ask a question and get answers grounded in your graph</text>
+<rect x="42" y="52" width="46" height="3" rx="1.5" fill="#a974ff"/>
+
+<g>
+<rect x="556" y="92" width="156" height="28" rx="14" fill="#19132b" stroke="#6b4fb0" stroke-width="1"/>
+<circle cx="574" cy="106" r="4" fill="#b98cff" filter="url(#glow)"/>
+<text x="588" y="110.5" font-size="12.5" font-family="ui-monospace, SFMono-Regular, Menlo, monospace" fill="#cdbcf2">cognee.recall()</text>
+</g>
+
+<g font-size="15">
+<rect x="120" y="100" width="67" height="23" rx="5" fill="#3a2566" opacity="0"><animate attributeName="opacity" values="0;0;1;1;0" keyTimes="0;0.10;0.18;0.86;1" dur="7s" repeatCount="indefinite"/></rect>
+<text x="52" y="116.5" fill="#9b93ab">Where</text>
+<text x="102" y="116.5" fill="#9b93ab">is</text>
+<text x="127" y="116.5" fill="#f3f0fa" font-weight="700">cognee</text>
+<text x="195" y="116.5" fill="#9b93ab">based?</text>
+</g>
+
+<g>
+<g stroke="#332d44" stroke-width="2" fill="none">
+<line x1="165" y1="322" x2="120" y2="272"/>
+<line x1="165" y1="322" x2="150" y2="374"/>
+</g>
+<circle cx="165" cy="322" r="12" fill="#564f6b"/>
+<circle cx="120" cy="272" r="9" fill="#433c57"/>
+<circle cx="150" cy="374" r="9" fill="#433c57"/>
+<text x="92" y="352" font-size="10.5" fill="#7d7590">existing memory</text>
+</g>
+
+<line x1="470" y1="302" x2="177" y2="322" stroke="#463a66" stroke-width="2"/>
+<line x1="410" y1="160" x2="535" y2="215" stroke="#463a66" stroke-width="2.5"/>
+<line x1="535" y1="215" x2="470" y2="302" stroke="#463a66" stroke-width="2.5"/>
+
+<line x1="535" y1="215" x2="470" y2="302" stroke="#b98cff" stroke-width="2.5" stroke-dasharray="108" stroke-dashoffset="108" filter="url(#glow)">
+<animate attributeName="stroke-dashoffset" values="108;108;0;0;108" keyTimes="0;0.36;0.5;0.86;1" dur="7s" repeatCount="indefinite"/>
+</line>
+<text x="505" y="262" font-size="11" fill="#c4a6ff" text-anchor="middle" opacity="0"><animate attributeName="opacity" values="0;0;1;1;0" keyTimes="0;0.42;0.5;0.86;1" dur="7s" repeatCount="indefinite"/>based in</text>
+
+<circle cx="410" cy="160" r="12" fill="#6b4fb0" filter="url(#softglow)"/>
+<text x="410" y="135" font-size="12.5" font-weight="700" fill="#c9c2da" text-anchor="middle">Vasilije</text>
+
+<circle cx="535" cy="215" r="22" fill="none" stroke="#b98cff" stroke-width="2" opacity="0"><animate attributeName="r" values="12;12;26;12;12" keyTimes="0;0.24;0.34;0.42;1" dur="7s" repeatCount="indefinite"/><animate attributeName="opacity" values="0;0;0.9;0;0" keyTimes="0;0.24;0.32;0.42;1" dur="7s" repeatCount="indefinite"/></circle>
+<circle cx="535" cy="215" r="13" fill="#a974ff" filter="url(#softglow)"/>
+<text x="535" y="190" font-size="12.5" font-weight="700" fill="#f3f0fa" text-anchor="middle">cognee</text>
+
+<circle cx="470" cy="302" r="22" fill="none" stroke="#d9bcff" stroke-width="2" opacity="0"><animate attributeName="r" values="13;13;13;28;13;13" keyTimes="0;0.5;0.52;0.62;0.7;1" dur="7s" repeatCount="indefinite"/><animate attributeName="opacity" values="0;0;0;0.9;0;0" keyTimes="0;0.5;0.52;0.6;0.7;1" dur="7s" repeatCount="indefinite"/></circle>
+<circle cx="470" cy="302" r="13" fill="#7a5cc0" filter="url(#softglow)"><animate attributeName="fill" values="#7a5cc0;#7a5cc0;#d9bcff;#d9bcff;#7a5cc0" keyTimes="0;0.52;0.6;0.86;1" dur="7s" repeatCount="indefinite"/></circle>
+<text x="470" y="339" font-size="12.5" font-weight="700" fill="#f3f0fa" text-anchor="middle">Berlin</text>
+
+<circle r="5" fill="#f0e4ff" filter="url(#glow)" opacity="0">
+<animate attributeName="cx" values="535;535;470" keyTimes="0;0.36;0.5" dur="7s" repeatCount="indefinite"/>
+<animate attributeName="cy" values="215;215;302" keyTimes="0;0.36;0.5" dur="7s" repeatCount="indefinite"/>
+<animate attributeName="opacity" values="0;0;1;1;0" keyTimes="0;0.36;0.42;0.5;0.54" dur="7s" repeatCount="indefinite"/>
+</circle>
+
+<g opacity="0"><animate attributeName="opacity" values="0;0;1;1;0" keyTimes="0;0.6;0.7;0.86;1" dur="7s" repeatCount="indefinite"/>
+<rect x="350" y="352" width="240" height="28" rx="14" fill="#19132b" stroke="#6b4fb0" stroke-width="1"/>
+<circle cx="368" cy="366" r="4" fill="#b98cff" filter="url(#glow)"/>
+<text x="382" y="370.5" font-size="13" fill="#cdbcf2">cognee is based in Berlin</text>
+</g>
+</svg>

cognee-1.2.2.dev2/assets/remember.svg

@@ -0,0 +1,69 @@
+<svg viewBox="0 0 740 400" xmlns="http://www.w3.org/2000/svg" role="img" font-family="ui-sans-serif, system-ui, -apple-system, Segoe UI, Roboto, Arial, sans-serif">
+<title>remember() turning a sentence into a memory graph</title>
+<desc>cognee.remember() reads "Vasilije founded cognee in Berlin", highlights three entities, which sprout as glowing purple nodes connected by labeled edges and link into existing memory.</desc>
+<defs>
+<filter id="glow" x="-80%" y="-80%" width="260%" height="260%"><feGaussianBlur stdDeviation="4" result="b"/><feMerge><feMergeNode in="b"/><feMergeNode in="SourceGraphic"/></feMerge></filter>
+<filter id="softglow" x="-120%" y="-120%" width="340%" height="340%"><feGaussianBlur stdDeviation="7" result="b"/><feMerge><feMergeNode in="b"/><feMergeNode in="SourceGraphic"/></feMerge></filter>
+</defs>
+
+<rect x="0" y="0" width="740" height="400" fill="#0a0a0d"/>
+
+<text x="40" y="42" font-size="21" font-weight="700" fill="#f3f0fa">Turn a sentence into a memory graph agents can reason over</text>
+<rect x="42" y="52" width="46" height="3" rx="1.5" fill="#a974ff"/>
+
+<g>
+<rect x="540" y="92" width="172" height="28" rx="14" fill="#19132b" stroke="#6b4fb0" stroke-width="1"/>
+<circle cx="558" cy="106" r="4" fill="#b98cff" filter="url(#glow)"/>
+<text x="572" y="110.5" font-size="12.5" font-family="ui-monospace, SFMono-Regular, Menlo, monospace" fill="#cdbcf2">cognee.remember()</text>
+</g>
+
+<g font-size="15">
+<rect x="45" y="100" width="66" height="23" rx="5" fill="#3a2566" opacity="0"><animate attributeName="opacity" values="0;1;1;0" keyTimes="0;0.10;0.86;1" dur="7s" repeatCount="indefinite"/></rect>
+<rect x="186" y="100" width="67" height="23" rx="5" fill="#3a2566" opacity="0"><animate attributeName="opacity" values="0;0;1;1;0" keyTimes="0;0.10;0.16;0.86;1" dur="7s" repeatCount="indefinite"/></rect>
+<rect x="282" y="100" width="55" height="23" rx="5" fill="#3a2566" opacity="0"><animate attributeName="opacity" values="0;0;1;1;0" keyTimes="0;0.16;0.22;0.86;1" dur="7s" repeatCount="indefinite"/></rect>
+<text x="52" y="116.5" fill="#f3f0fa" font-weight="700">Vasilije</text>
+<text x="119" y="116.5" fill="#9b93ab">founded</text>
+<text x="193" y="116.5" fill="#f3f0fa" font-weight="700">cognee</text>
+<text x="261" y="116.5" fill="#9b93ab">in</text>
+<text x="289" y="116.5" fill="#f3f0fa" font-weight="700">Berlin</text>
+</g>
+
+<g>
+<g stroke="#332d44" stroke-width="2" fill="none">
+<line x1="165" y1="322" x2="120" y2="272"/>
+<line x1="165" y1="322" x2="150" y2="374"/>
+</g>
+<circle cx="165" cy="322" r="12" fill="#564f6b"/>
+<circle cx="120" cy="272" r="9" fill="#433c57"/>
+<circle cx="150" cy="374" r="9" fill="#433c57"/>
+<text x="92" y="352" font-size="10.5" fill="#7d7590">existing memory</text>
+</g>
+
+<line x1="470" y1="302" x2="177" y2="322" stroke="#a974ff" stroke-width="2" stroke-dasharray="296" stroke-dashoffset="296" filter="url(#glow)">
+<animate attributeName="stroke-dashoffset" values="296;296;0;0;296" keyTimes="0;0.47;0.57;0.86;1" dur="7s" repeatCount="indefinite"/>
+</line>
+
+<line x1="410" y1="160" x2="535" y2="215" stroke="#a974ff" stroke-width="2.5" stroke-dasharray="140" stroke-dashoffset="140" filter="url(#glow)">
+<animate attributeName="stroke-dashoffset" values="140;140;0;0;140" keyTimes="0;0.27;0.37;0.86;1" dur="7s" repeatCount="indefinite"/>
+</line>
+<line x1="535" y1="215" x2="470" y2="302" stroke="#a974ff" stroke-width="2.5" stroke-dasharray="115" stroke-dashoffset="115" filter="url(#glow)">
+<animate attributeName="stroke-dashoffset" values="115;115;0;0;115" keyTimes="0;0.37;0.47;0.86;1" dur="7s" repeatCount="indefinite"/>
+</line>
+
+<g font-size="11" fill="#c4a6ff" opacity="0"><animate attributeName="opacity" values="0;0;1;1;0" keyTimes="0;0.37;0.45;0.86;1" dur="7s" repeatCount="indefinite"/>
+<text x="455" y="181" text-anchor="middle">founded</text></g>
+<g font-size="11" fill="#c4a6ff" opacity="0"><animate attributeName="opacity" values="0;0;1;1;0" keyTimes="0;0.47;0.55;0.86;1" dur="7s" repeatCount="indefinite"/>
+<text x="505" y="262" text-anchor="middle">based in</text></g>
+
+<circle cx="410" cy="160" r="0" fill="#a974ff" filter="url(#softglow)"><animate attributeName="r" values="0;0;16;14;14;0" keyTimes="0;0.143;0.22;0.27;0.86;1" dur="7s" repeatCount="indefinite"/></circle>
+<g opacity="0"><animate attributeName="opacity" values="0;0;1;1;0" keyTimes="0;0.24;0.30;0.86;1" dur="7s" repeatCount="indefinite"/>
+<text x="410" y="135" font-size="12.5" font-weight="700" fill="#f3f0fa" text-anchor="middle">Vasilije</text></g>
+
+<circle cx="535" cy="215" r="0" fill="#a974ff" filter="url(#softglow)"><animate attributeName="r" values="0;0;16;14;14;0" keyTimes="0;0.243;0.32;0.37;0.86;1" dur="7s" repeatCount="indefinite"/></circle>
+<g opacity="0"><animate attributeName="opacity" values="0;0;1;1;0" keyTimes="0;0.34;0.40;0.86;1" dur="7s" repeatCount="indefinite"/>
+<text x="535" y="190" font-size="12.5" font-weight="700" fill="#f3f0fa" text-anchor="middle">cognee</text></g>
+
+<circle cx="470" cy="302" r="0" fill="#c9a8ff" filter="url(#softglow)"><animate attributeName="r" values="0;0;16;14;14;0" keyTimes="0;0.343;0.42;0.47;0.86;1" dur="7s" repeatCount="indefinite"/></circle>
+<g opacity="0"><animate attributeName="opacity" values="0;0;1;1;0" keyTimes="0;0.44;0.50;0.86;1" dur="7s" repeatCount="indefinite"/>
+<text x="470" y="339" font-size="12.5" font-weight="700" fill="#f3f0fa" text-anchor="middle">Berlin</text></g>
+</svg>

{cognee-1.2.2.dev0 → cognee-1.2.2.dev2}/cognee/api/v1/improve/improve.py

@@ -40,6 +40,7 @@ async def improve(
     node_name: Optional[List[str]] = None,
     session_ids: Optional[List[str]] = None,
     build_global_context_index: bool = False,
+    build_truth_subspace: bool = False,
     **kwargs: Unpack[ImproveKwargs],
 ):
     """Enrich an existing knowledge graph with additional context and rules.
@@ -86,6 +87,10 @@ async def improve(
             context index after default enrichment. Skipped in background
             mode because ordered background pipeline chaining is not
             supported yet.
+        build_truth_subspace: Opt-in flag (default ``False``) for building the
+            truth subspace from distilled session learnings after distillation
+            and before enrichment. Only runs when ``session_ids`` is provided.
+            Off by default = no behaviour change.
         **kwargs: Additional options -- see ``ImproveKwargs``.
 
     Returns:
@@ -113,6 +118,7 @@ async def improve(
             "session_ids": ",".join(session_ids) if session_ids else "",
             "run_in_background": run_in_background,
             "build_global_context_index": build_global_context_index,
+            "build_truth_subspace": build_truth_subspace,
             "cognee_version": cognee_version,
         },
     )
@@ -199,6 +205,26 @@ async def improve(
                 if distilled:
                     stages_run.append("distill_sessions")
 
+                # Stage 2d: build the truth subspace from distilled session
+                # learnings (opt-in, default OFF). Runs after distillation so
+                # freshly accepted lessons are available as anchors, and before
+                # enrichment. Non-fatal — never blocks the rest of improve().
+                if build_truth_subspace:
+                    try:
+                        from cognee.modules.truth_subspace.build import (
+                            build_truth_subspace as _build_truth_subspace,
+                        )
+
+                        result_ts = await _build_truth_subspace(
+                            dataset=dataset,
+                            session_ids=session_ids,
+                            user=user,
+                        )
+                        logger.info("improve: truth subspace built -> %s", result_ts)
+                        stages_run.append("build_truth_subspace")
+                    except Exception as e:
+                        logger.warning("improve: truth subspace build failed (non-fatal): %s", e)
+
             # Stage 3: default enrichment (triplet embeddings)
             from cognee.modules.memify import memify
 

{cognee-1.2.2.dev0 → cognee-1.2.2.dev2}/cognee/api/v1/recall/recall.py

@@ -5,6 +5,7 @@ from uuid import UUID
 from pydantic import BaseModel, Field
 from typing_extensions import TypedDict
 
+from cognee.base_config import get_base_config
 from cognee.context_global_variables import set_session_user_context_variable
 from cognee.exceptions import CogneeValidationError
 from cognee.infrastructure.databases.vector.embeddings.config import EmbeddingConfig
@@ -375,7 +376,7 @@ async def recall(
     context_profile: str = "qa",
     wide_search_top_k: int | None = 100,
     triplet_distance_penalty: float | None = 6.5,
-    feedback_influence: float = 0.0,
+    feedback_influence: float = get_base_config().default_feedback_influence,
     verbose: bool = False,
     retriever_specific_config: dict | None = None,
     neighborhood_depth: int | None = None,

{cognee-1.2.2.dev0 → cognee-1.2.2.dev2}/cognee/api/v1/search/search.py

@@ -8,6 +8,7 @@ from cognee.infrastructure.databases.vector.embeddings.config import EmbeddingCo
 from cognee.infrastructure.llm.config import LLMConfig
 from cognee.modules.search.types import SearchResult, SearchType
 from cognee.modules.users.methods import get_default_user
+from cognee.base_config import get_base_config
 from cognee.modules.search.methods import search as search_function
 from cognee.modules.data.methods import get_authorized_existing_datasets
 from cognee.modules.data.exceptions import DatasetNotFoundError
@@ -43,7 +44,7 @@ async def search(
     session_id: Optional[str] = None,
     wide_search_top_k: Optional[int] = 100,
     triplet_distance_penalty: Optional[float] = 6.5,
-    feedback_influence: float = 0.0,
+    feedback_influence: float = get_base_config().default_feedback_influence,
     verbose: bool = False,
     retriever_specific_config: Optional[dict] = None,
     neighborhood_depth: Optional[int] = None,

{cognee-1.2.2.dev0 → cognee-1.2.2.dev2}/cognee/base_config.py

@@ -14,6 +14,9 @@ class BaseConfig(BaseSettings):
     cache_root_directory: str = get_absolute_path(".cognee_cache")
     logs_root_directory: str = os.getenv("COGNEE_LOGS_DIR", str(Path.home() / ".cognee" / "logs"))
     monitoring_tool: object = Observer.NONE
+    # Default blend weight for the learned feedback signal during graph search.
+    # Opt-in by default to preserve existing retrieval behavior.
+    default_feedback_influence: float = float(os.getenv("DEFAULT_FEEDBACK_INFLUENCE", "0.0"))
 
     @pydantic.model_validator(mode="after")
     def validate_paths(self):