Free
$0/mo
$10/day in credits, ad-free
Amp
Autonomous agentic coding by Sourcegraph — terminal-first, multi-model, no markup.
Amp is Sourcegraph's autonomous agentic coding tool — the successor to Cody Free and Cody Pro. It runs in the terminal and editor extensions, executes complex multi-file coding tasks using multiple frontier models (Claude, GPT-5, Gemini), and supports parallel subagents with independent context windows. Freemium with a $10/day free credit grant and pay-as-you-go at cost-price (no markup) beyond that.
+Pros
- Multi-model architecture that dynamically uses GPT-5.5, Opus 4.7, and fast models — each deployed for what it does best, rather than a single-model approach.
- Three built-in agent modes (deep, smart, rush) let you dial between extended reasoning, unconstrained capability, and fast/cheap execution depending on task complexity.
- Extensible TypeScript plugin API with lifecycle hooks (tool.call, tool.result, agent.start/end), custom commands, custom tools, and UI interactions — inspired by Pi's extension system.
- Subagents and Oracle tools enable parallelized complex work: subagents handle independent subtasks in separate context windows, while Oracle (GPT-5.5) provides a powerful second opinion for debugging and code review.
- Built-in Librarian subagent can search all public GitHub code plus your private repos, enabling cross-repository research and deep library/framework understanding without leaving the CLI.
- Thread sharing with version-control-like visibility levels (public, unlisted, workspace, private), searchable by keyword/file/author, and referencable via @-mentions between threads — essentially git for agent conversations.
−Cons
- Closed-source proprietary product — the core Amp CLI has no public source repository, creating full vendor lock-in to the Sourcegraph/AmpCode ecosystem.
- Aggressively opinionated development philosophy that frequently removes features the team doesn't personally love (Amp Tab, Fork command, TODO lists, Custom Commands, editor extensions all killed).
- No built-in approval prompts before executing tools by default — Amp runs shell commands and edits files automatically, which is a security risk in untrusted repositories without manual plugin configuration.
- Pricing is opaque for heavy usage — execute mode (amp -x) consumes only paid credits (free tier is interactive-only), and smart/deep modes use expensive frontier models that can scale costs significantly.
Pricing
Pay-as-you-go
At cost
Zero-markup model cost pricing beyond free grant
Enterprise
Custom
Sourcegraph batch changes, deep search
Amp is Sourcegraph's autonomous agentic coding tool, launched in 2025 as the successor to Cody Free and Cody Pro after those plans were discontinued in July 2025. Where Cody was primarily an IDE assistant with chat and inline suggestions, Amp is built around fully autonomous, multi-step task execution — closer to an agent you delegate work to than a tool you converse with. It runs natively in the terminal and offers VS Code extensions, making it accessible in any workflow that includes a CLI.
Amp's standout architecture feature is parallel subagents: for large or complex tasks, Amp can spawn multiple independent subagents each with their own context window, working on different parts of the codebase simultaneously, then merge the results. It also integrates an 'oracle' — GPT-5 — which the agent autonomously invokes for particularly complex debugging or reasoning tasks, selecting the right model for the right subtask without developer intervention. Threads allow developers to save, revisit, and share complete agentic sessions for collaboration and review.
Pricing is designed to be fully transparent: Amp Free provides $10/day in model credits (ad-free since March 2026), and beyond that, usage is billed at model cost price with zero markup — what Sourcegraph pays the model providers is exactly what users pay. Sourcegraph also integrates Amp into the broader Sourcegraph Enterprise platform, which provides Sourcegraph's Search API codebase context layer, batch changes, code monitoring, and MCP server capabilities for teams that need organizational-scale codebase understanding.
Key Features
• Successor to Cody Free & Cody Pro (discontinued July 23, 2025) — displaced users received Amp credits
• CLI-first autonomous agent: complex multi-file coding tasks from the terminal
• Parallel subagents: spawns independent agents with separate context windows for large tasks
• Oracle (GPT-5) invocation: agent autonomously selects GPT-5 for complex reasoning/debugging subtasks
• Multi-model: Claude, GPT-4o, GPT-5, Gemini — model selection based on task suitability
• Threads: save, revisit, and share complete agentic sessions for collaboration and review
• VS Code extension + VS Code-compatible forks (Cursor, Windsurf)
• Sourcegraph Search API context: cross-repo codebase awareness when used within Sourcegraph Enterprise
• MCP support via Sourcegraph platform
• Free: $10/day in credits, ad-free (since March 2026)
• Pay-as-you-go: zero-markup model cost pricing beyond the free grant
• Enterprise: custom, integrates with Sourcegraph batch changes, code monitoring, deep search
Version History
Amp Neo is now just Amp. The rebuilt CLI is available to everyone. Removed the --take-me-back flag. Infrastructure stability and performance improvements.
Added passkey authentication that can be required for sensitive Amp operations, adding a security layer for destructive or production-affecting actions.
Rush mode now runs on GPT-5.5 with no reasoning — twice as fast, more interactive, and tuned for small well-defined tasks.
The npm package for the Amp CLI is now @ampcode/cli (migrated from @sourcegraph/amp).
Claude Opus 4.7 is now powering smart mode. It solves harder problems but vague prompts get weaker results.
Signature Snippet
A developer runs in the terminal: 'amp: Migrate all API endpoints from Express 4 to Hono, update the OpenAPI schema, and write integration tests for each route.' Amp spawns parallel subagents with separate context windows — one per service — executes the migration, merges the results, runs the tests, and presents a summary diff. The session is saved as a Thread and shared with the team lead for review.More in this Space
Prime Agent (Prime Intellect)
Prime Agent is an open-source (MIT) self-improving coding and research harness by Prime Intellect, built on a Recursive Language Model that runs inside a persistent IPython REPL and a Continual Harness that refines its own state via /refine. It took the same frontier weights from 30.2% to 95.5% on ARC-AGI-3 RHAE Best@1, powered daemon-backed sessions, and ships no hosted offering.
OpenCodeReview (ocr)
OpenCodeReview (ocr) is Alibaba's open-source AI code-review CLI. It reads git diffs and drives an LLM agent through a hybrid architecture where deterministic pipelines handle file selection, rule matching, and comment positioning while the agent does the reasoning. Built on two years of internal use at Alibaba scale, it ships a multi-language ruleset and consumes roughly 1/9 of the tokens of general-purpose agents.
Delta (by Zed)
Multiplayer environment from Zed for coding with agents. Threads unify the agent conversation, the code edits, and human review in one shareable artifact, replacing the branch/PR loop. Built on DeltaDB, a CRDT-based version control layer that records every edit between commits alongside the prompts and reasoning that produced them.