production ready

Continuously fresh context for AI.

Turn codebases, meeting notes, inboxes, videos … into live context for your agents to reason over effectively — with minimal incremental processing. Fresh data anytime.

Star on GitHub—
Incremental · only the delta Any scale · parallel by default Declarative · Python, 5 min
codebase · real-time Δ only cocoindex.flow live SRC/ page.tsx layout.tsx header.tsx utils.ts db.ts api.ts call_graph(file) · who calls whom, rebuilt incrementally as imports change CALL GRAPH main init load cfg api HIERARCHY ▾ app ▾ ui Button Card api route VECTORS SYMBOLS indexCode fn Chunker cls embed fn flow mod sink fn 2,410 FILES 18,074 CHUNKS Δ 12
Built with CocoIndex

CocoIndex-code · Super-charge coding agent with live context.

Call graphs, hierarchies, symbol tables, and semantic indexes — all kept fresh as the repo changes. Learn more about CocoIndex Code →

Star on GitHub—
YOUR REPO src/ flow.py pipeline.py embed.py api/ route.ts utils.ts ON CHANGE Δ CALL GRAPH main embed SYMBOLS embed fn Chunker cls sink mod VECTORS CHUNKS · AST CONTEXT LIVE · FRESH CODING AGENT ▸ where is embed() called? ◯ 3 callers · freshly indexed src/flow.py:42 src/pipeline.py:18 tests/t_embed.py:7 ▸
Δ +1

Incremental processing

Only the delta is reindexed. Sub-second freshness at any repo size.

Index & semantic search

Less grep. Find by meaning — functions, patterns, intent — not string matches.

Call graphs & blast radius

Know exactly what a change touches before it ships. Trace every caller and callee.

Global view

Spot duplicates. Understand architecture across the whole repo, not one file.

Build your own

Coding agents

Generate · refactor

Code-review agents

Catch · approve

Security-review agents

Scan · audit

Built with CocoIndex

Built with CocoIndex. Let's go!

Working starters. Clone, plug your source, ship. Each one is a handful of files and a flow declaration.

Codebase · real-time def index_code (repo): # walk the tree for file in repo.files: chunks = chunk(file) emb = embed(chunks) sink(emb, "postgres") @cocoindex.flow Fork This Idea →

Real-time codebase indexing

Keep an index of your repo in sync with every commit. Feed code-review and coding agents with structure, not raw text.
Meetings · knowledge graph person:nora topic:q2 action:ship doc:rfc-14 meeting Fork This Idea →

Meeting notes → knowledge graph

Extract people, topics, decisions and action items from notes into a live graph. Query them with your agent.
Hacker News · trends t-6h t-3h now ↑ agents ↑ rust Fork This Idea →

HN trending topics detector

Ingest Hacker News incrementally. Detect surfacing topics before they peak. Perfect for weekend hacking.
See all examples →
How it works

CocoIndex is an incremental engine for long-horizon agents.

Data transformation for any engineer, designed for AI workloads — with a smart incremental engine for always-fresh, explainable data.

Learn the concept →

Python Native Transformation Codebases Meeting Notes Web · APIs File System · Blob Stores Databases Message Queues Images · Video Voice · Transcripts Relational DB Data Warehouse Vector DB Graph DB Message Queue Feature Store Source file · walk_dir() yields FileLike — this run's input FileLike await coco.map(process_chunk, splitter.split(text)) split() → chunks embed(chunk) → vector · memoized (skipped when input+code unchanged) embed(chunk) → vec embed(chunk) → vector · memoized embed(chunk) → vec Same path, content changed → new fingerprint; downstream re-runs only for delta chunks FileLike (Δ) splitter.split(text) re-runs on the changed file; unchanged chunks stay memoized downstream split() re-runs @coco.fn(memo=True) · input unchanged → embed skipped, cached vector reused cache hit · no re-run delta detected · input fingerprint changed → await embed(chunk) re-runs Δ → re-embed
CocoInsight LIVE Lineage src sink Observability CocoIndex Persistent Data Pipeline Control Plane LIVE Caching Reuse what hasn't changed — only the delta runs. Pipeline Catalog Every flow registered — find, fork, reuse. Version Tracking Code, schema, data — all versioned end-to-end. Continuously Learning The engine adapts as your data and code evolve. Lineage Every byte in the target traces back to a source. Task Scheduling Parallel by default — low-latency, low-cost. Metrics Collection Throughput, freshness, cost — all observable. Failure Management Retries, back-off, dead letters — no data loss.
Built for agents

Reliable. Autonomous. Minimalistic.

Agents break when their data lies. CocoIndex makes the data tell the truth — through every source change, every code change, and every long-running job to back long horizon agents.

Δ only Δ Data change

Source data changed. We noticed. Before you did.

When source changes

One file edited → one row re-syncs.

Don't think about it. The framework watches the source, computes the delta, and reconciles the target — at any scale, in parallel.

Incremental by default
v2 live Code change

Code changed. Schema auto evolved. No migration meeting.

When F changes

Ship new code → only affected rows re-run.

Your target store is already connected to live agents? No worries. Only changed code gets rerun. Schemas evolve automatically.

No index swap · no downtime
The mental model

React — for data engineering.

A persistent-state-driven model. You declare the desired state of your target. The engine keeps it in sync with the latest source data and code, across long time horizons, with low latency and low cost.

Your code is as simple as the one-off version.

See the React ↔ CocoIndex mental model →

Target = F ( Source )

TARGET F · YOUR CODE @coco.fn process(src) SOURCE a.py b.md c.pdf d.ts
ENGINE · AUTO-SYNC · Δ ONLY
def

Python, not a DAG.

You write the transform. The engine derives the graph.

Declare target state.

We compute the minimum work to reach it.

Lineage end-to-end.

Every byte in the target traces to a source.

Δ +1

Incremental at any scale.

Only the delta runs — never the full recompute.

CocoInsight

What's going on with my data?

Step-by-step, understand what your pipeline is doing. Think of it as real-time rendering in the browser — for data.

CocoInsight

Every step. Every record. Explainable.

See the shape of your data at every stage of the flow. Trace a single vector back to the paragraph it came from. Debug with your eyes, not grep.

Flow · pdf_ingestlive · 24 rec/s
SOURCE CHUNK EMBED SINK rfc-14.pdf readme.md spec.pdf cached new.pdf e5-mistral · 1024d · 68% postgres · table: docs 5 files · 1 new 142 chunks · Δ 8 142 vec · Δ 8 upserted · 8
● running · reused 94% cocoinsight.local
Vibing?

Vibe-coding native. Pipeline ready in 5 min.

Describe the flow. Claude writes the cocoindex. You run it. The framework keeps it fresh forever.

Try the Claude skill →
youindex my /docs folder into Postgres.
claudewiring cocoindex.flow · chunk → embed → sink…
okflow.py · 14 lines · running
log[00:12] 142 chunks · 1024d · 68% cached
Loved by builders

Incredible optimizations, out of the box.

I'm in love with CocoIndex. ❤️ It's a very mature project — with incredible optimizations like incremental processing, parallel chunking, and maximum efficiency built right in. These are hard to design and maintain, yet they just work out of the box.

I'm inspired to learn Rust because I want to contribute to CocoIndex and Zed. Both represent the best of engineering excellence and community spirit.

And honestly — CocoIndex has one of the most responsible, thoughtful communities I've seen.

Shivansh Subramanian · Startup Founder

Your agents deserve fresh context.

Get your agent ready to production in 10 min with reliable and fresh data.

Star on GitHub Learn the Concept