The Gap in Cross-Session Memory
When asking AI to research or read source code, closing the conversation window means starting from scratch next time—this is a common pain point with current AI coding assistants. Researchers must repeatedly have the model re-search the same information, re-read the same code files, wasting time and tokens.
ascent-research is an open-source project that allows AI research work to persist and accumulate across sessions. All research findings are saved as plain text files locally, and the next session automatically picks up where the last one left off.
Core Design
ascent-research works as follows:
- Local Plain Text Storage: All research findings, code analyses, and web summaries are saved as
.txtfiles in a local directory, with no dependency on cloud databases - Automatic Context Recovery: On startup, it scans existing research files and injects previous findings into the current session’s context
- Multi-Source Support: Can collect information from web pages, local codebases, Obsidian notes, and other sources
- Progress Visualization: Research progress is displayed intuitively through file structure—showing which modules have been analyzed and which need deeper exploration
The core idea is to externalize “research state” as a file system, rather than binding it to a particular conversation instance’s context window.
Practical Workflow
# 1. Initialize a research project
ascent-research init my-project --source-dir ./src --web
# 2. Let AI start researching
# AI analyzes the specified directory's code structure and generates a research report
# Web URLs can also be added as supplementary references
# 3. Restart after closing the conversation
ascent-research resume my-project
# Automatically loads previous research findings, AI continues from last progress
# 4. Export research results
# All findings are saved as plain text, viewable, editable, or shareable with the teamComparison with Alternatives
| Dimension | Traditional Conversation | ascent-research |
|---|---|---|
| Cross-Session | No memory | Plain text persistence |
| Storage Method | Platform-bound | Local file system |
| Context Recovery | Manual pasting | Automatic loading |
| Auditability | Not exportable | Plain text directly viewable |
| Collaboration | Single-person conversation | Shareable research directory |
Use Cases and Limitations
Suitable for:
- Progressive analysis and understanding of large codebases
- Technical research requiring multiple iterative rounds
- Team sharing of research results
- Research needs that avoid uploading code to the cloud
Limitations:
- Plain text format doesn’t include rich interaction history (e.g., code execution result screenshots)
- Requires manual management of research project lifecycle
- Limited help for scenarios requiring real-time responsiveness (e.g., online debugging)
- Project is in early stages with a small community
ascent-research addresses a specific but high-frequency pain point: the fragmentation of AI research work. It doesn’t attempt to replace conversational AI, but rather provides a persistent infrastructure layer for cross-session research.