Cloud Parts

Augment Code

A commercial coding agent for VS Code, JetBrains and the terminal built around a Context Engine that keeps a live semantic index of your codebase. On 8 October 2026 Harness bought the Context Engine, the Auggie CLI and the Cosmos cloud agents.

What it is
AI coding agents (editor agents)
Type
Commercial product
Made by
Augment Code (Augment Computing, Inc.)
License
Proprietary
Facts checked
9 October 2026
Website

What it is

Augment’s difference is retrieval. When you open a workspace, the extension uploads your code to Augment’s cloud and its Context Engine builds a search index for you alone: an embedding index, kept per developer so it follows your branch, that Augment says updates within seconds of a change. Embeddings are computed and searched on Augment’s own infrastructure rather than through third-party APIs, and a proof-of-possession check means you can only retrieve code you actually have locally. .gitignore and .augmentignore decide what gets indexed. The agent in VS Code and JetBrains (chat and completions also exist for Vim and Neovim) uses that index to pull the relevant code even from monorepos of more than 100,000 files, then edits files, runs terminal commands, calls MCP servers and native integrations such as GitHub, Jira and Linear, and keeps checkpoints you can roll back to.

Auggie, the CLI launched in September 2025, puts the same agent and index in the terminal: an interactive TUI, auggie --print for scripts and CI, an ACP mode for Zed, Neovim and Emacs, and an MCP server mode that hands the Context Engine to other agents such as Claude Code, Codex or Cursor. Cosmos runs agents in cloud or self-hosted sandboxes, triggered by tickets, pull requests or alerts. You choose the model (Claude, GPT, Gemini, Kimi, GLM and others) and pay the provider’s list price plus a 40% service fee. On 8 October 2026 Harness bought Cosmos, the Auggie CLI, the Code Context Engine and related technology, and the team behind them moved to Harness. The VS Code and JetBrains extensions were still getting releases when the deal was announced.

Why teams pick it

  • Retrieval built for large codebases: a per-developer index that follows branch switches within seconds, across several repositories
  • The Context Engine is usable outside Augment: Auggie’s MCP server mode gives other agents its codebase search
  • Enterprise controls: SOC 2 Type II, ISO/IEC 42001, customer-managed keys, and no training on customer code
  • One agent across VS Code, JetBrains, the terminal, CI and cloud sandboxes

Watch out for

  • Ownership is in flux: Harness bought the Context Engine, the Auggie CLI and Cosmos in October 2026, and the announcements say nothing about the future of the IDE extensions
  • Indexing uploads your code to Augment’s cloud, and retrieval runs there rather than on your machine
  • Closed source and paid only, with a 40% fee on top of model list prices; pricing moved from per-message plans to credits in October 2025 and then to token-based billing
  • Tool permission rules are enforced by the CLI and Cosmos but not by the IDE extension
  • Its published quality gains, such as 30 to 80% better results for agents using the Context Engine MCP, come from Augment’s own benchmarks

Alternatives

What all editor agents have in common →
ClineOpen sourceAn open-source coding agent that started as a VS Code extension and now runs the same engine in JetBrains, a terminal CLI, a desktop app and an SDK, with whatever model you bring.+1 Roo Code

How it fits with other technologies

Auggie in a script, and the Context Engine plugged into Claude Code over MCP · bash

npm install -g @augmentcode/auggie
auggie login                       # opens a browser to sign in

# One-shot from the repository root: indexes it, answers, prints to stdout
auggie --print "How do requests reach the billing service?"

# Give another agent Augment's codebase-retrieval tool
claude mcp add-json auggie --scope user \
  '{"type":"stdio","command":"auggie","args":["--mcp","--mcp-auto-workspace"]}'

What it costs (checked 2026-10-09)#

Augment bills in dollars of usage. Every request costs the model’s tokens at the provider’s public list price, plus a service fee, plus sandbox time when Cosmos runs it.

PriceNotes
Standard$20 a month$20 of usage included, up to 50 seats, no per-seat charge
Business$100 a month$100 of usage included, up to 50 seats
Service fee40% of LLM spendFunds the Context Engine and Cosmos; no fee on compute
Cosmos compute$0.19 per hourBilled in 5-minute increments, rounded up
EnterpriseCustomSSO, SCIM, customer-managed keys, unlimited users
  • From Augment’s pricing page and token-based pricing docs, checked on 2026-10-09, the day after the Harness deal was announced.

See it in a real stack

No architecture uses Augment Code yet. Drop it into the playground and wire it up yourself.Open playground →
Where it came from2024–2026

Augment was founded by Igor Ostrovsky, formerly chief architect at Pure Storage and an engineer at Microsoft, and Guy Gur-Ari, an AI researcher from Google, with Scott Dietzen as CEO. It came out of stealth on 24 April 2024 with a $227 million Series B at a $977 million valuation, $252 million in total. Its pitch was that existing assistants didn’t understand a programmer’s intent or a team’s code. It built its own indexing and retrieval rather than calling generic embedding APIs.

The product grew from completions and chat into Augment Agent for VS Code and JetBrains in April 2025, with memories, MCP and native integrations, and then the Auggie CLI in September 2025. Heavy agent use broke its pricing: in October 2025 Augment said one user on the $250 Max plan was costing it close to $15,000 a month, and moved from per-message plans to credits, and later to token-based billing. In February 2026 it opened the Context Engine to other agents over MCP, and it built Cosmos to run fleets of agents from tickets, reviews and alerts.

On 8 October 2026 Harness, a software delivery platform company, announced it had acquired select Augment Code assets: Cosmos, the Auggie CLI, the Code Context Engine and related technology. The team behind them, including co-founder and CTO Igor Ostrovsky, joined Harness, and Cosmos became the Harness Cosmos Software Factory Agent.

Why it’s built this way

Build its own indexing and embedding search
Augment argued that generic embedding APIs plus a hosted vector database gave poor quality, slow updates and a security risk, since embeddings can be reversed into code. Running its own models on its own infrastructure let it update a per-developer index within seconds.
One index per developer, not per repository
Developers switch branches and rebase all the time. Retrieving from the main branch, or from an index refreshed every 10 minutes, feeds the model names that don’t exist in the code being edited.
Proof of possession for retrieval
A shared cloud index could leak code to people who shouldn’t see it. Requiring the client to prove it holds the code avoids managing per-file permissions.
Expose the Context Engine to other agents
Many teams already used Claude Code, Cursor or Codex. Offering the index over MCP let those agents use Augment’s retrieval, which Augment said improved their results by 30 to 80% in its own tests.

Timeline

  1. 2024

    Emerges from stealth on April 24 with a $227 million Series B

  2. 2025

    Augment Agent for VS Code and JetBrains on April 2

  3. 2025

    Auggie CLI in September; credit-based pricing from October 20

  4. 2026

    Context Engine MCP for any agent on February 6

  5. 2026

    Harness acquires Cosmos, Auggie and the Context Engine on October 8

Sources17 · facts checked 9 October 2026

Improve this page on GitHub