NEW/v1/insights/movements — exact metric-movement decomposition→

Git-native org index · SQL + Semantic

The deepest indexof your GitHub org.

Every commit, diff, rename and owner — cloned straight from git, not scraped file-by-file from the API — plus the PRs and reviews only the forge holds. One index you query with SQL or semantic search in a single call, verified complete against git itself.

See what you can build

13endpoints

7MCP tools

7SQL views

$0to start

Depth of index

Most tools index files.
We index the whole history.

Code search stops at the current tree. EOS clones the whole object store — every commit, diff and rename, not a file-by-file API crawl — then precomputes the relationships nobody else stores: ownership decay, temporal coupling, review flow, expertise.

9entity types modeled — commits, PRs, reviews, files, authors, teams, deps, decisions, releases

100%of history cloned on install and verified complete against git's own count — then kept fresh on every push

2query languages over one index — read-only SQL and cited semantic search

0uncited answers — every field carries the age of the data behind it

Proven at scale

The whole of VS Code,
indexed in one pass.

microsoft/vscode — 116,608 commits and 30,388 files across all of history — cloned and indexed end to end on a single machine. After the first pass, every push is incremental: only the delta, never a re-index.

116,608commits indexed — the entire history of microsoft/vscode, not just the current tree

~26 minone cold pass on a single machine — then incremental on every push

1.98Mlines blamed to their author across the full history

7languages parsed for symbols — the L3 layer, built on top in the same pass

What you build on it

Any view, one query away.

01 / 08

CHORD · TEMPORAL COUPLING

Blast radius before you touch it

Files that move together in the same PR, even with no import between them. The edge an import graph cannot see.

GET /v1/context/graph/couplingmcp · get_coupling · Starter+

GET /v1/context/graph/coupling
  ?repo=acme/api&min_shared=3

Get connected · two minutes

Install → index → ask.

eos.dev eos.dev/docs

01Install the GitHub AppPick the repos. EOS backfills PRs, commits, reviews and contributors, indexes the code, and keeps it fresh on every push.

02Grab an API keySettings → API. Keys look like eos_… and every response carries its own freshness stamp.

03Point your agent at itMCP for agents, plain HTTP for everything else. No SDK required — the API is the product.

What are you building with? · Claude Code

terminalclaude code · mcp

# install the MCP server

npm i -g @eos-ai/mcp-server

# point Claude Code at your index

claude mcp add eos-context --env EOS_API_KEY=eos_your_key -- eos-mcp

# then just ask

claude "who should review PR 482, and what breaks if I touch services/auth?"

Same index behind Claude Desktop, Cursor, Windsurf and plain HTTP. Keys and the full tool list live at eos.dev/docs.

REST

GET /v1/context/ownership

?path=services/auth

MCP server

npm i -g @eos-ai/mcp-server

EOS_API_KEY · Claude, Cursor, Windsurf

SQLSTARTER

POST /v1/context/query

select repo, count(*) from pull_requests

Grounded answers

Cited and dated, not a vibe.

CitationsSearch answers cite the file and lines they came from — your agent can check instead of trusting.

FreshnessEvery response says how old its data is (meta.freshness). If ingest is behind, the API says so instead of guessing.

Exact numbersInsight math decomposes: a lead-time movement is attributed per team and per shift of work — terms sum to the delta, no interaction fudge.

Why EOS

Why not just the GitHub API?

EOS runs on GitHub — the App and webhooks feed it the PRs and reviews only the forge holds. What it doesn't do is rebuild your org from REST calls on every request. It clones the object store once, then tracks the delta.

GitHubEOS

SourcingFile-by-file REST crawlClone once, fetch the delta

Deletionsremoved[] webhook, usually unwiredgit diff — always correct

RenamesCapped at 1,000No cap

CompletenessPartial index still reports "ready"Verified against git's count

History searchKeywordsSemantic, cited

Limits5k/hr, org-widePer-key quota

Free$05,000 requests / mo
1 indexed repo · 25 agent answers
Ownership, contributors, reviewers, search

Starter$5$50 billed annually50,000 / mo, then $2 per 1k
3 indexed repos, then $1 / repo / mo
Adds graph, coupling, imports, activity, read-only SQL and webhooks

EnterpriseTalk to usannual contractCommitted volume
SSO / SAML · audit logs
Self-hosting · GHES · GitLab

The paid boundary is the code index, not endpoint walls. Free covers the core read surface — and one indexed repo, so indexing your own repo costs nothing. Starter unlocks the index-derived surfaces we pay embeddings to build; beyond it, scale is metered, not tiered.

Index once.
Let every agent read it.

Free to start, no card.