The durable workspace for agentic research.
Bring project context, prompts, agent reports, and syntheses into one browsable knowledgebase. Bamba keeps the source Markdown in Cloudflare R2 and indexes the research structure in D1, so your work stays easy to inspect, import, and reuse.
Markdown-first · Multi-folder import · Encrypted vault · Hashed shares · Cloudflare-native
Import with control
From folders of agent outputs to research records.
Drop in one or more topic folders containing Markdown reports. Bamba stages files locally first, runs a fail-closed preflight against the same limits the server enforces, lets you review titles and inferred authors, and only uploads when you confirm.
- 1. Stage locallySelect or drop multiple topic folders into the workspace.
- 2. Preflight, then reviewCatch size, count, and collision issues before upload—then check topic names, report counts, and inferred authors.
- 3. Save to the workspaceStore the Markdown in R2 and the searchable structure in D1.
3 Markdown reports staged
Keep the source reports
Prompts, reports, and syntheses you can still read.
Bamba distinguishes report kinds—prompt, report, and synthesis—so the research record stays structured while every document remains inspectable Markdown. Move between tabs without losing the brief, and correct author/model labels when filenames are ambiguous.
- Prompt, report, and synthesis kinds stay labeled in the topic reader.
- Report Markdown remains readable as Markdown, not trapped in chat history.
- Author/model labels can be corrected and reused across imports.
Architecture Evaluation
The decoupled R2 storage ensures long-term immutability of agent evidence documents without vendor database lock-in.
By separating object storage from relation indexes, Bamba maintains lightweight, zero-latency metadata queries.
Markdown-first architecture
Files you can inspect. Structure you can browse.
Bamba stores project, topic, prompt, and report Markdown as Cloudflare R2 objects while D1 indexes the relationships and metadata used by the app. The result is a research library built around durable files instead of opaque conversation history.
projects/kodeverdict/
├── README.md
└── topics/
├── 1-arbitration/
│ ├── topic.md (Prompt Brief)
│ └── reports/
│ ├── claude.md
│ ├── qwen.md
│ └── gpt.md
└── 2-consensus/
├── topic.md
└── reports/
└── synthesis.mdSmarter ingestion
Teach Bamba how your agents name files.
Create simple filename rules that assign an author or model during batch import. A file containing “claude,” “qwen,” or another pattern can be tagged automatically, with the option to correct it later.
Explore workspace →"claude"→"qwen"→"gemini"→Encrypted Research Vault
Project-scoped artifacts, sealed with AES-256-GCM.
Drop supporting files—including reusable .md resources—into the project vault. Packages are compressed, then envelope-encrypted with AES-256-GCM before durable storage, scoped to the project so sensitive attachments stay alongside the research without mixing into the inspectable report library.
- AES-256-GCM envelopes bound to owner and resource scope.
- Any file type, including .md resources such as synthesis prompts.
- Kept separate from the browsable Markdown report trail.
Compressed client-side, encrypted before storage, scoped to this project.
Selective sharing
Hashed share links for individual reports.
Mint a public link for a single report when you need outside review. Bamba stores a hash of the token, supports expiry and revocation, and keeps the rest of the workspace private.
- Per-report links—not whole-project exposure.
- Token hashed at rest; revoke or rotate when access should end.
- Recipients see the shared Markdown report, nothing more.