docs(readme): reposition around auto-learning#115
Conversation
Reframes the README from "shared memory / agent amnesia" toward
auto-learning: trace capture -> skill codification -> skill propagation
-> compounding capability. Tagline kept ("One brain for all your
agents"). No technical surface changes: HIVEMIND_* env vars, install
commands, agent list, embeddings/summaries/skilify CLI tables,
architecture, and security all intact.
Changes:
- Subtitle: "shared memory" -> "auto-learning, cloud-backed shared brain"
- Pull quote: agent-amnesia -> Monday->Tuesday migration framing
- Lead paragraph reframed around traces / codified skills / propagated
capability
- Feature bullets: Captures, Codifies, Searches, Propagates, Intercepts,
Summarizes (was: Captures, Searches, Shares, Intercepts, Summarizes)
- "How it works": replaced 3-block diagram with 4-stage pipeline (trace
capture, skill codification, skill propagation, compounding capability)
plus larger diagram
- Natural search examples mention traces and codified skills
- Team sharing: "see your memory" -> "draw from your team's traces and
skills"
- Data collection notice: added "Codified skills" row and workspace-
isolation framing
- Skills (skilify) intro: framed as "where auto-learning becomes concrete"
- New Roadmap section before Security: trajectory export for fine-tuning,
dense vector retrieval, skill versioning/review, more agents
|
Claude finished @kaghni's task in 1m 45s —— View job Review
Two accuracy regressions and one verifiability concern flagged inline:
The language pass otherwise reads well and the technical surface is intact as you noted. |
Coverage ReportNo Generated for commit 36ed9cf. |
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
| ## Roadmap | ||
|
|
||
| - **Trajectory export for fine-tuning.** Because traces are stored in Deeplake's tensor format, they're export-ready as PyTorch datasets. Teams running their own open-source models can fine-tune on their org's accumulated trajectories. A handful of advanced customers are already doing this against the trajectories their Claude Code and Codex agents generated. | ||
| - **Dense vector retrieval.** Currently using BM25 + grep fallback. Dense retrieval over GPU-accelerated vector search is in active development. |
There was a problem hiding this comment.
"Currently using BM25 + grep fallback" contradicts the "Optional: enable semantic search (embeddings)" section above (lines 298–348), which documents a working nomic-embed-text-v1.5 daemon producing 768-dim message_embedding / summary_embedding columns and hybrid semantic+lexical ranking. Dense retrieval already exists — it's just opt-in. The roadmap item should be GPU-accelerated / scaled dense retrieval, not "dense retrieval" as if none ships today.
| - **Dense vector retrieval.** Currently using BM25 + grep fallback. Dense retrieval over GPU-accelerated vector search is in active development. | |
| - **GPU-accelerated dense retrieval at scale.** Hivemind already supports local CPU embeddings via the optional nomic-embed daemon (see "Optional: enable semantic search" above). Roadmap: GPU-accelerated vector search over the full trace store, on by default. |
|
|
||
| ## Roadmap | ||
|
|
||
| - **Trajectory export for fine-tuning.** Because traces are stored in Deeplake's tensor format, they're export-ready as PyTorch datasets. Teams running their own open-source models can fine-tune on their org's accumulated trajectories. A handful of advanced customers are already doing this against the trajectories their Claude Code and Codex agents generated. |
There was a problem hiding this comment.
Worth confirming before this ships publicly — "A handful of advanced customers are already doing this against the trajectories their Claude Code and Codex agents generated" is a specific, falsifiable claim about real customer usage. If this is the growth team's framing rather than a verified fact, consider softening (e.g. "designed so teams running their own open-source models can fine-tune on their org's accumulated trajectories") or dropping the sentence — public README claims about customer behavior tend to attract scrutiny.
|
Important Review skippedAuto incremental reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Plus Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
📝 WalkthroughWalkthroughREADME.md updated with reframed product positioning as an "auto-learning, cloud-backed shared brain," restructured "How it works" into four explicit phases, generalized skill-codification language away from Claude-Code specifics, refreshed examples and phrasing, expanded data collection transparency disclosures, and added a new Roadmap section listing trajectory export, vector retrieval, skill versioning, and additional agent support. ChangesDocumentation Positioning and Content Refresh
🎯 2 (Simple) | ⏱️ ~10 minutes
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Tip 💬 Introducing Slack Agent: The best way for teams to turn conversations into code.Slack Agent is built on CodeRabbit's deep understanding of your code, so your team can collaborate across the entire SDLC without losing context.
Built for teams:
One agent for your entire SDLC. Right inside Slack. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@README.md`:
- Around line 239-244: The fenced code block containing the quoted lines ("What
was Emanuele working on?" / "Search traces for authentication bugs we've solved"
/ "What did we decide about the API design?" / "Show me skills my team has
codified for handling migrations") must include a language specifier so it
renders correctly; update that triple-backtick block to use a language
identifier (for example add "text" after the opening ```), ensuring the rest of
the block content is unchanged.
- Around line 202-229: The fenced ASCII diagram block in README.md lacks a
language specifier; change the opening fence from ``` to include a language
identifier (e.g., use ```text) so the ASCII diagram renders correctly—update the
code block that contains the box diagram (the multi-line ASCII art starting with
the top border "┌─────────────────────────────────────────────────────┐") to
start with ```text.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
| ``` | ||
| ┌─────────────────────────────────────────────────────┐ | ||
| │ Your Coding Agent │ | ||
| │ Your Coding Agents │ | ||
| │ Claude Code · Codex · OpenClaw · Cursor · ... │ | ||
| └──────────────────────────┬──────────────────────────┘ | ||
| │ | ||
| ┌──────────────────▼──────────────────┐ | ||
| │ 📥 Capture (every turn) │ | ||
| │ 📥 Trace capture (every turn) │ | ||
| │ prompts · tool calls · responses │ | ||
| └──────────────────┬──────────────────┘ | ||
| │ | ||
| ┌──────────────────▼──────────────────┐ | ||
| │ 🧠 Hivemind │ | ||
| │ SQL tables · Virtual File System │ | ||
| │ Search Memory · inject context │ | ||
| │ 🧠 Skill codification │ | ||
| │ pattern detection · LLM extraction │ | ||
| │ workspace-scoped │ | ||
| └──────────────────┬──────────────────┘ | ||
| │ | ||
| ┌──────────────────▼──────────────────┐ | ||
| │ 🔗 Skill propagation │ | ||
| │ injected into agent context │ | ||
| │ every agent · every teammate │ | ||
| └──────────────────┬──────────────────┘ | ||
| │ | ||
| ┌──────────────────▼──────────────────┐ | ||
| │ 🌊 Deeplake │ | ||
| │ Shared across all agents │ | ||
| │ Postgres · S3 │ | ||
| │ Tensor storage · Postgres · S3 │ | ||
| └─────────────────────────────────────┘ | ||
| ``` |
There was a problem hiding this comment.
Add language specifier to fenced code block.
The ASCII diagram should specify a language identifier for proper rendering.
📝 Proposed fix
-```
+```text
┌─────────────────────────────────────────────────────┐
│ Your Coding Agents │📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| ``` | |
| ┌─────────────────────────────────────────────────────┐ | |
| │ Your Coding Agent │ | |
| │ Your Coding Agents │ | |
| │ Claude Code · Codex · OpenClaw · Cursor · ... │ | |
| └──────────────────────────┬──────────────────────────┘ | |
| │ | |
| ┌──────────────────▼──────────────────┐ | |
| │ 📥 Capture (every turn) │ | |
| │ 📥 Trace capture (every turn) │ | |
| │ prompts · tool calls · responses │ | |
| └──────────────────┬──────────────────┘ | |
| │ | |
| ┌──────────────────▼──────────────────┐ | |
| │ 🧠 Hivemind │ | |
| │ SQL tables · Virtual File System │ | |
| │ Search Memory · inject context │ | |
| │ 🧠 Skill codification │ | |
| │ pattern detection · LLM extraction │ | |
| │ workspace-scoped │ | |
| └──────────────────┬──────────────────┘ | |
| │ | |
| ┌──────────────────▼──────────────────┐ | |
| │ 🔗 Skill propagation │ | |
| │ injected into agent context │ | |
| │ every agent · every teammate │ | |
| └──────────────────┬──────────────────┘ | |
| │ | |
| ┌──────────────────▼──────────────────┐ | |
| │ 🌊 Deeplake │ | |
| │ Shared across all agents │ | |
| │ Postgres · S3 │ | |
| │ Tensor storage · Postgres · S3 │ | |
| └─────────────────────────────────────┘ | |
| ``` |
🧰 Tools
🪛 markdownlint-cli2 (0.22.1)
[warning] 202-202: Fenced code blocks should have a language specified
(MD040, fenced-code-language)
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@README.md` around lines 202 - 229, The fenced ASCII diagram block in
README.md lacks a language specifier; change the opening fence from ``` to
include a language identifier (e.g., use ```text) so the ASCII diagram renders
correctly—update the code block that contains the box diagram (the multi-line
ASCII art starting with the top border
"┌─────────────────────────────────────────────────────┐") to start with
```text.
| ``` | ||
| "What was Emanuele working on?" | ||
| "Search memory for authentication bugs" | ||
| "Search traces for authentication bugs we've solved" | ||
| "What did we decide about the API design?" | ||
| "Show me skills my team has codified for handling migrations" | ||
| ``` |
There was a problem hiding this comment.
Add language specifier to fenced code block.
The code example should specify a language identifier for proper rendering.
📝 Proposed fix
-```
+```text
"What was Emanuele working on?"
"Search traces for authentication bugs we've solved"🧰 Tools
🪛 markdownlint-cli2 (0.22.1)
[warning] 239-239: Fenced code blocks should have a language specified
(MD040, fenced-code-language)
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@README.md` around lines 239 - 244, The fenced code block containing the
quoted lines ("What was Emanuele working on?" / "Search traces for
authentication bugs we've solved" / "What did we decide about the API design?" /
"Show me skills my team has codified for handling migrations") must include a
language specifier so it renders correctly; update that triple-backtick block to
use a language identifier (for example add "text" after the opening ```),
ensuring the rest of the block content is unchanged.
Cuts the README from 643 lines to 322 (~50% reduction), in line with peer projects (mem0 ~250, supermemory 362, letta 122, mastra 82). Detail moves to dedicated files under docs/ that the README now points to. Dedup cuts: - Lead paragraph: drop "Repeat work stops being repeat work..." marketing fluff (the auto-learning thesis is already in the bullets and How-it-works) - Quick start: drop the parenthetical agent list (duplicates the Supported- assistants table 20 lines later) - How it works: drop the 4-step narrative paragraphs and ASCII diagram; collapse to a one-paragraph "Capture -> Codify -> Propagate -> Compound" summary. The bullets above already convey the same flow - Features: cut the AI-summaries and Team-sharing teaser subsections (already covered by the Summaries H2 and the Data-collection notice) - Skilify intro: drop the "this is where auto-learning becomes concrete" editorial opener Extract to docs/: - docs/EMBEDDINGS.md - semantic-search daemon, install/uninstall, lexical fallback (~50 inline lines moved) - docs/SUMMARIES.md - wiki-worker triggers, generation flow, env-var reference (~64 inline lines moved) - docs/SKILIFY.md - skilify worker, pull/unpull, gate-CLI per agent, configuration, logs (~150 inline lines moved) - docs/ARCHITECTURE.md - integration model per agent + monorepo tree (~36 inline lines moved) Each section now has a one-paragraph summary + link to the full guide. Modernization (Tier 1): - Add npm-version and GitHub-stars badges to the badge row - Drop the ASCII flow diagram (modern peer READMEs use prose + image, not ASCII art) No technical surface changes: HIVEMIND_* env vars, install commands, agent list, and all command references are intact (just relocated).
Local CPU embeddings already ship via the optional nomic-embed daemon (documented in the Semantic search section and docs/EMBEDDINGS.md), so "currently using BM25 + grep fallback" was wrong. The actual roadmap item is GPU-accelerated dense retrieval at scale, on by default. Caught by claude[bot] PR review.
Wrap the white logo paths in a #141214 background rect (matches the landing-page dark surface) and scale the paths to 70% so they don't touch the edges. Without this the logo was invisible on light/white backdrops (GitHub light theme, Google SERP, browser tabs).
Five prose mentions in README plus the docs/SKILIFY.md file (created in this PR) renamed to docs/SKILLIFY.md to match. Source-code rename is landing in a separate PR.
There was a problem hiding this comment.
I'd leave it in the README
| hivemind skilify unpull --all # ALSO remove flat-layout (locally-mined) skills — destructive | ||
| hivemind skilify unpull --legacy-cleanup # ALSO remove pre-`--author`-layout `<projectkey>/` dirs from older skilify versions | ||
| hivemind skillify # show current scope, team, install, per-project state | ||
| hivemind skillify scope <me|team|org> # who counts as "in scope" for mining |
There was a problem hiding this comment.
remove it. it's enough in the specific md
2 participants
Summary
Reframes the README from "shared memory / agent amnesia" toward auto-learning: trace capture → skill codification → skill propagation → compounding capability. Tagline kept ("One brain for all your agents").
Source: growth team's positioning draft (
SUGGENSTED_README.md, kept locally and not committed). We pulled the language shifts and rejected the technical regressions in their version (renamed env vars, dropped the unifiedhivemind installflow, simplified per-agent install to manualgit clone, single-agent hook lifecycle table).Changes
"Persistent, cloud-backed shared memory"→"Auto-learning, cloud-backed shared brain""see your memory"→"draw from your team's traces and skills"Codified skillsrow and workspace-isolation framingNot changed
All technical surface intact: env vars, the unified
npm install -g @deeplake/hivemind && hivemind installquick start, per-agent install/uninstall commands, agent-detail collapsibles, embeddings section, summaries env-var table, skilify CLI/triggers/gate-CLI tables, integration-model architecture table, monorepo structure, Security, Development.Test plan
#skills-skilifyfrom the new How-it-works section, etc.)Summary by CodeRabbit