# The coding agent that understands **Source:** [Codistry](https://codistry.ai) --- Codistry maps your entire codebase, so every change carries through to every caller, test, and dependency.Codistry maps your entire codebase and learns how you work. Every change carries through to every caller, test, and dependency: the parts other agents skip. Codistry in VS Code, citing the code it read ## Your codebase, explained. Your tokens, saved. Codistry explains your codebase once, then hands the model only the code each task actually needs. Same work, fewer tokens, with any model you run. kept current as you commit · [how the context engine works](https://codistry.ai/docs/context-engine)·[we measured the cost](https://codistry.ai/blog/what-does-an-ai-coding-agent-actually-cost) one task through the context engineatlas-api · local index the ask Why do OAuth users get signed out when their API keys rotate? the engine checks its map match: Auth Sessions · files that change together when keys rotate the pack · handed to the model - rotate-keys.tsmints the replacement key - oauth-session.tsowns the session cache - session.test.tspins the expected behavior what the model reads the repo, re-explained the pack half the tokens, measured ## A coding assistant that learns how you work Codistry adapts to your codebase and your habits with every session. It writes in your conventions, remembers what you've explained, and picks up where you left off. It maps your expertise from your commit history alone: it defers to you where you lead, and explains more where you're new. Codistry · Expertisemined from git history the Expertise view, built from your commits · [how expertise works](https://codistry.ai/docs/expertise) ## Any model you want, whenever you want it Codistry isn't tied to one lab. Use a cheap model for the routine work and a frontier model for the hard problems, switching from one prompt to the next. Bring your own keys, run local models, or use hosted models on your plan. As new models ship, you move when you want to. Everything Codistry learns about your code lives in your editor and your repo. Cloud models see only what you choose to send, and with local models it runs fully offline, even air-gapped. one session · three prompts, three models “Bump the deps and fix the lint fallout”Claude Haiku “Find why sessions expire after key rotation”Claude Opus “Explain the auth module to a new hire”Ollama Claude Opusfrontier Ollamalocal · air-gap ready Claude Haikufast + cheap ClaudeGPTCodistry hostedOpenRouterOllamaLM Studioyour endpoint switch mid-session, nothing to reconfigure · [supported models](https://codistry.ai/docs/models)·[what the extension sends](https://codistry.ai/docs/telemetry) ## Works with the stack you already have Codistry runs in VS Code, alongside the agents and tools your team already uses. Bring it in for the hard parts where other assistants get lost, and leave the rest of your workflow exactly as it is. ### Context engine Maps your repo into a knowledge graph: symbols, relationships, and how the code changes together over time. Kept current as you commit. Watch one build: ready · watching commits98 symbols · 144 relationships · 6 areas ### Parallel sessions Run several tasks side by side. Switching between them never cancels a run. auth refactorrunning fix flaky CI testneeds input db migrationdone ### Subagents Hand a self-contained chunk of work to an isolated agent with its own brief and tools. Only the result comes back. your session report mergedsubagent · research how auth works ### MCP integrations Hook up databases, APIs, and internal tools over the MCP protocol. Pre-built servers or your own. PostgresGitHubFigmayour API ### Sandboxed skills Specialized skills run inside isolated Docker containers, away from the rest of your system. $skill: generate-pdf-report → isolated container · no host access done · your system untouched ## Three steps to your first session No procurement, no pilot program. [See pricing →](https://codistry.ai/pricing) 1. ### Install the extension Grab it from the VS Code marketplace, or email yourself an install link to open at your desk. 2. ### Connect a model Use hosted models on the Developer plan, paste your own API key, or point it at a local model through Ollama or LM Studio. 3. ### Hand it a real task Open a repo and ask for an actual change, not a toy demo. ### No idea what to hand it? By the time the index is built, Codistry has ideas of its own: work drawn from your recent commits and the areas you own. You won't start from a blank prompt. Codistry · New session drawn from your recent commits, refreshed as you work ## Try it on your own codebase Pick a plan, open a project, and give Codistry something real to do. The first session is the demo.