Skip to content
Appvecta

Appvecta Agent

Bring your own model. Keep the approval.

The agent is one interface with several adapters behind it. Appvecta is not trying to sell you inference — the Brain, Run, Device Hub and QA all work with no AI provider configured at all.

Planned for the private alpha

Provider-neutral

ProviderHow it authenticates
Claude CodeYour installed, already-logged-in CLI, in its official non-interactive mode
Codex CLIThe same contract, via its supported streaming interface
API-compatibleAnthropic Messages, OpenAI-compatible (OpenAI, OpenRouter, Gemini-compatible)
Ollama / localA local endpoint; degrades to plan-only mode if the model cannot call tools
Appvecta-hostedLater, as one more provider behind the same interface

The subscription rule. Appvecta uses an existing subscription only through the provider’s own officially supported login or CLI. It never scrapes browser cookies, copies website session tokens, imitates private consumer endpoints, or asks you to paste browser credentials.

Use Appvecta from the agent you already use

The engine will run a local MCP server that exposes the Brain as tools. Any MCP-capable tool — Claude Code, Codex, Cursor and others — can use them with your own account, from outside Appvecta. These work with no AI provider configured in Appvecta at all.

brain.search        brain.definition     brain.callers
brain.dependencies  brain.featurePath    brain.impact
brain.contextPack

How a request becomes context

Classify the intent → query the Brain → retrieve direct symbols plus one or two relationship levels → add relevant configuration and tests → include exact source ranges, not whole folders → state what is missing or low-confidence → let the agent ask for more → log every file and graph fact supplied.

Every context pack records the tokens sent alongside a baseline — the whole files a naive agent would have read. The interface shows the difference. The engineering target is a pack at most 25% of that baseline.

Plan, approve, apply, revert

  • Before anything is written, the plan shows the goal, planned files, direct and indirect impact, database, API and build risks, tests to add or run, and the commands requested.
  • After the change you get the diff, build and test results, QA evidence and any unresolved warnings.
  • "Revert this task" restores only that task’s hunks and leaves your unrelated edits intact.
  • There are no silent autonomous edits, in any edition.

Agent tools are typed and allowlisted

The allowlist does not depend on what a model or a repository asks for. The only Git commands available to the agent are diff, status, log and show.

searchFiles      readRange         queryBrain
requestSubgraph  proposePatch      applyPatch        (approval)
gitDiff          runCommand        (category-gated)
readCommandResult  requestQaPlan   launchQa          (approval)
readReport       replayFlow
All featuresNext: Appvecta QA