Augment Code
Also known as: Auggie CLI, Context Engine, Cosmos
Augment Code is an AI coding assistant evolving into an AI-native engineering platform. Cosmos (June 2026) adds teams of specialized agents that coordinate across the full software development lifecycle with shared memory, MCP and webhook integrations, and multi-agent orchestration.
Augment Code is an AI coding platform aimed at large, complex codebases and at engineering organizations rather than individual developers. At its center is a proprietary Context Engine that semantically maps a codebase by structure, tracking what calls what, what is active, and what is deprecated across repositories, libraries, documentation, and history.
That engine feeds a set of agents that behave less like autocomplete and more like a senior teammate. In VS Code and JetBrains, the agent turns a natural-language task into edits and pull requests, working through multi-step plans and pausing for approval before it runs terminal commands.
A terminal tool, the Auggie CLI, brings the same capability to the command line and to CI pipelines in a headless mode, and a first-party review bot reviews pull requests with full codebase awareness.
Cosmos runs teams of Experts, reusable agents with their own environment, model and memory, that wake on GitHub, Linear, Slack, PagerDuty, cron or webhook events and hand work to one another, with template Experts for PR authoring, review, incident investigation and ticket management.
Users choose among models from Anthropic, Google, OpenAI, xAI and others, and the Context Engine supplies only the slice of context a task needs. Because the Context Engine is exposed over the Model Context Protocol and an SDK, its codebase understanding can also be plugged into other agents and clients, and the platform connects to tools like Jira, Linear, Slack, and GitHub.
The product is built for regulated, security-conscious organizations. It carries SOC 2 Type II and ISO/IEC 42001 certification, does not train on customer code, and offers customer-managed encryption, VPC, single-tenant and on-premises options, sandboxed agent execution, single sign-on, role-based access, and audit logging. Pricing is flat per team with usage included rather than per seat. Augment's focus is squarely on the problems that appear when many engineers and many agents work in the same large system: architectural reasoning, shared knowledge, and consistency across services.
Vendor details
Canonical URL
https://www.augmentcode.com/
Category
Coding agent
Company status
independent
Use cases & customers
Target customers
Deployment options
Integrations
As an MCP client, Augment connects GitHub, Linear, Jira and any MCP compatible tool, and the Context Engine itself is exposed as a standalone MCP server that other agents including Claude Code, Cursor, Codex and Zed can call for codebase retrieval, in a local stdio mode or a hosted cross repo mode indexing default branches via a GitHub App. First party GitHub Actions ship for PR review and PR description, ACP support connects Zed, JetBrains, Neovim and Emacs, the CLI is published on npm as @augmentcode/auggie, and enterprise service accounts issue dedicated API tokens for CI and automation.
In practice
Your payment system spans hundreds of thousands of files across dozens of services, and ordinary assistants hallucinate on it. Augment's Context Engine maps the whole codebase semantically, so its agent reasons about cross-service impact instead of guessing from one file.
A refactor needs to stay consistent across seventeen files without breaking callers. Augment's IDE agent plans the change, applies it with full dependency awareness, and opens a pull request you can review.
You're in a regulated org where code can't train external models or leave your perimeter. Augment carries SOC 2 and ISO 42001 certification, never trains on your code, and offers VPC deployment with customer-managed keys.
Sources & related URLs
Related / legacy domains
Agentic Index coverage score
12.0 / 14 capabilities · 86%
| Integrations & Tool Calling | Full |
|---|---|
|
Experts and the Auggie CLI call external tools through MCP servers and native tools, Cosmos connects to GitHub, Linear, Slack, PagerDuty and custom webhooks, and tool permissions allow, deny or delegate each tool call to a webhook or script policy. Sourcedocs.augmentcode.com/cosmos/getting-started, cli/permissions and augmentcode.com/pricingread 2026-09-29 |
|
| Workflow Orchestration | Full |
|
Cosmos runs multiple Experts side by side, each doing its part, hands work off through workers, subagents and Expert to Expert delegation, and ships a Project Builder that turns a large feature into a design, a ticket breakdown and coordinated implementation, with a Ticket Dispatcher sending ready work to implementation workers. Sourceaugmentcode.com/product/cosmos and docs.augmentcode.com/cosmos/experts-templatesread 2026-09-29 |
|
| Knowledge Grounding & RAG | Full |
|
The Context Engine semantically indexes and maps code across hundreds of thousands of files and retrieves curated context from dependencies, documentation, recent changes, commit history and issues, served to Augment's own agents and to other agents through an MCP server in local or hosted mode and a Context Engine SDK. Sourceaugmentcode.com/context-engine and docs.augmentcode.com/context-services/mcp/overviewread 2026-09-29 |
|
| Human Oversight & Guardrails | Full |
|
By default the IDE agent pauses before it runs a terminal command or reaches an external integration, and the user reviews the suggested action and approves it before work continues; checkpoints allow reverting, CLI tool permissions allow, deny or delegate calls to a webhook or script policy, and Cosmos states that every action that matters is approved by a human. Sourcedocs.augmentcode.com/using-augment/agent, cli/permissions and augmentcode.com/product/cosmosread 2026-09-29 |
|
| Security, Identity & Governance | Full |
|
Augment holds SOC 2 Type II attestation and ISO/IEC 42001 certification, with HIPAA and a BAA available, no training on customer code, and customer controlled encryption. Enterprise adds SSO, OIDC and SCIM and granular access controls, Cosmos adds granular RBAC and CMEK, and service accounts act as non human identities with their own token lifecycles. Sourceaugmentcode.com/security, augmentcode.com/pricing and augmentcode.com/product/cosmosread 2026-09-29 |
|
| Observability & Auditability | Full |
|
Each Cosmos session records every message, agent turn and tool call, is kept without expiry and can be reopened, shared across the organization or restricted, and Cosmos adds audit logs that feed a SIEM and replayable runs, so a buyer can inspect what an Expert did step by step. Sourcedocs.augmentcode.com/cosmos/sessions-overview and augmentcode.com/product/cosmosread 2026-09-29 |
|
| Memory & State Persistence | Partial |
|
Expert memory is captured by the Expert itself, stored as readable Markdown in the organization's virtual filesystem, scoped per repository or per global, channel, project or user scope, and loaded at the start of relevant work, with a noisy memory model that promotes learning only once evidence is sufficient. No lifetime, expiry or purge path is documented for Expert memory. Sourcedocs.augmentcode.com/cosmos/experts-memory and experts-code-review-memoryread 2026-09-29 |
|
| Deployment & Data Residency | Full |
|
Cosmos names VPC deployment, single tenant instances and on premises options, and lets a team run the daemon and CLI on its own machines exposed to Cosmos; Enterprise adds data residency options and multi region compute. Sourceaugmentcode.com/product/cosmos and augmentcode.com/pricingread 2026-09-29 |
|
| Prebuilt Agents, Templates & Packs | Full |
|
Augment ships fourteen Template Experts that each do their own job, including PR Author, PR Fixer, Deep Reviewer, Pair Reviewer, Risk Analyzer, Verifier, Incident Investigator, Feedback Triager, Project Builder, Ticket Manager, Ticket Dispatcher and Data Analyst. Custom slash commands and published examples add reusable workflows. Sourcedocs.augmentcode.com/cosmos/experts-templates and cli/custom-commands-examplesread 2026-09-29 |
|
| Triggers & Channel Coverage | Full |
|
Cosmos triggers turn events into Expert actions without a person starting them: GitHub pull requests, Linear status changes, Slack messages, PagerDuty incidents, cron schedules and custom webhooks start the right Expert in a sandboxed VM. Sourcedocs.augmentcode.com/cosmos/getting-started and augmentcode.com/product/cosmosread 2026-09-29 |
|
| Model Flexibility & Routing | Full |
|
The user chooses among models from Anthropic, Google, OpenAI, xAI, Zhipu AI and Moonshot AI, per Expert or session in Cosmos and with /model or --model in the Auggie CLI. There is no documented way to bring your own model or key. Sourcedocs.augmentcode.com/models/available-modelsread 2026-09-29 |
|
| APIs, SDKs & MCP Extensibility | Full |
|
The Auggie SDK in TypeScript and Python builds custom integrations and agents on Auggie with typed and streaming output and custom tools, the Context Engine SDK ships an API reference, and auggie cloud manages Cosmos Experts, environments and sessions programmatically. Sourcedocs.augmentcode.com/cli/sdk and context-services/sdk/api-referenceread 2026-09-29 |
|
| Testing, Debugging & Optimization | Partial |
|
A Verifier Expert exercises a change against a running environment and reports evidence backed findings, and Cosmos mentions replayable runs. These check the customer's code and the agent's output, and no harness to evaluate or regression test agent behavior is documented. Sourcedocs.augmentcode.com/cosmos/experts-templates and augmentcode.com/product/cosmosread 2026-09-29 |
|
| Browser & Computer Use | Not documented |
|
No browser control or operation of software through a graphical interface is documented. Experts and the CLI work through the IDE, terminal, repositories, sandboxed VMs, CI and MCP connected tools. Sourceaugmentcode.com/product/cosmos and docs.augmentcode.comread 2026-09-29 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Recent platform changes
Augment Code released Cosmos Week 30, introducing session forking to branch conversations from completed turns and cost analytics with detailed charts for enterprise organizations. The release also expanded the Model Context Protocol (MCP) catalog with Google Workspace integrations and added Claude Fable 5 to the available model picker.
Bears on: Observability / auditability
View sourceAugment Code released Auggie CLI 0.33.0, introducing cloud triggers to enable or disable persistent workflow triggers and a worker sessions view to display child worker processes. The update also improves Model Context Protocol (MCP) reliability with proactive usability probing, automated secret scrubbing, and dynamic tool list refreshing.
Bears on: Workflow orchestration
View sourceAugment Code added Claude Fable 5 to the model picker, positioning it for long, multi-step, deep-reasoning work. This is a dated release event; current Fable 5 availability has not been separately verified.
Bears on: Agent capability
View sourcePricing
From $20 per month flat (Standard, up to 50 seats, $20 of usage included)
hybrid
Included quota
Standard includes $20 and Business $100 of monthly usage across LLM, Context Engine and compute, pooled across up to 50 seats; Enterprise usage is custom.
What is public
Augment publishes three plans: Standard at $20 per month and Business at $100 per month, both flat for up to 50 seats with the same dollar amount of usage included, and custom Enterprise. Usage covers LLM inference at the provider's rates plus a flat 40 percent fee on LLM usage (no fee on compute) and Cosmos compute time, pooled across the team, with pay as you go top ups.
Billing mechanics
A flat monthly plan price includes an equal amount of usage, pooled across up to 50 seats with no per seat charge. Usage is metered as LLM inference at provider rates plus a 40 percent fee on LLM usage and Cosmos compute time, and top ups are bought pay as you go once the included usage runs out.
Cost watchouts
The 40 percent fee on LLM usage and Cosmos compute time sit on top of provider model rates, so heavy agent use costs more than the plan price.
Variable cost rationale
Included usage is a fixed dollar amount and everything beyond it is metered by model usage and compute time, so spend depends on how much the agents run.
Additional watchouts
Cost follows usage rather than seats: model choice and task size drive LLM spend, the 40 percent fee applies to all LLM usage, and Cosmos compute time is billed on top.
Overage / add-ons
Usage beyond the included amount is bought as pay as you go top ups, metered as LLM inference at provider rates plus a flat 40 percent fee on LLM usage and Cosmos compute time; Enterprise usage is custom.
Sales call required
Mixed (some tiers require a call)
Free / trial
No free tier; the pricing page mentions trials without publishing their terms.
Lowest paid plan
Standard at $20 per month flat for up to 50 seats, with $20 of usage included.
Commercial notes
Enterprise adds custom user pricing, unlimited users, SSO, OIDC and SCIM, CMEK and ISO 42001 compliance, data residency options, multi region compute, granular access controls, audit trails and dedicated support.
Key ambiguities
The plan price is a floor: actual cost depends on LLM usage at provider rates plus the 40 percent fee and on Cosmos compute time.
Cancellation / refund
Standard and Business are self serve monthly plans; Enterprise is custom contracted.
Support SLA / resale
Community support on Standard and Business, where the SLA does not cover trials or beta use; Enterprise adds dedicated support.
Missing data
Trial terms are not published, and per task usage is not given as a benchmark.
Related vendors
- Cognition — Maker of Devin, an autonomous AI software engineer
- 10Web — Agentic website platform whose specialized AI agents build, host,…
- AgentUI — Managed AI app builder for operations teams that generates and hosts…
- Aider — Open source, model agnostic terminal coding agent that edits your…
- Anthropic Claude Code — Anthropic's agentic coding system across terminal, desktop, IDE, web…
- AppFactor — Agentic platform that runs persistent agents across an enterprise…
Alternatives to Augment Code
The closest documented capability profiles to Augment Code among coding agents tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Factory12.5 / 14Adds documented Browser & Computer UseAugment Code vs Factory →
- Tabnine11.5 / 14A lighter documented profile than Augment CodeAugment Code vs Tabnine →
- Cognition13.0 / 14Adds documented Browser & Computer UseAugment Code vs Cognition →
- Poolside11.0 / 14A lighter documented profile than Augment Code
- Warp13.0 / 14Adds documented Browser & Computer Use
- Zencoder13.0 / 14Adds documented Browser & Computer UseAugment Code vs Zencoder →
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded