Skip to main content
ChatCLI’s DEVIN provider drives the local devin binary in non-interactive mode and uses it purely as a transport to reach the LLM behind it. Everything else — conversation, context attachment, memory, compaction, sessions, the agent/coder tool protocol — stays in ChatCLI.
Why a wrapper? In enterprise deployments Devin is customized by Cognition (API + IDE integrations) and the HTTP API is not documented. The CLI is the supported surface and carries its own SSO authentication (devin auth login) — ChatCLI never speaks the private protocol, so Cognition-side changes are absorbed by their CLI, not by you.

​
Setup

  1. Install the Devin CLI and authenticate once (corporate SSO):
  1. That’s it — ChatCLI registers the provider automatically when the binary resolves (from DEVIN_CLI_PATH, from PATH, or from the well-known install directories):
No API key, no OAuth setup in ChatCLI: authentication belongs to the devin binary. Without the binary the provider simply doesn’t appear — same UX as a provider without credentials.
IDE-spawned servers (ACP/MCP) see a minimal PATH. Editors launch chatcli acp / chatcli mcp-server with the GUI-session environment — on macOS that PATH carries no Homebrew or npm directories, which used to make DEVIN vanish from those servers while the terminal REPL listed it fine. When the PATH lookup misses, ChatCLI now also probes the standard install locations (~/.local/bin, ~/bin, ~/.devin/bin, /opt/homebrew/bin, /usr/local/bin, Linuxbrew; on Windows %LOCALAPPDATA%\Programs\devin, %APPDATA%\npm, scoop shims), so the provider stays available without per-IDE env plumbing. An explicit DEVIN_CLI_PATH always wins and never falls back — and fixing the env at runtime followed by /reload recovers the provider without restarting.
Upgrading the Devin CLI does not break the provider. From the 3000.6 line onward the CLI refuses to run non-interactively in a directory it has not been told to trust, and --print has no way to raise the trust prompt. Every ChatCLI turn — and the model listing — runs in a private temp directory created moments earlier that holds nothing but the prompt file, so ChatCLI passes --respect-workspace-trust false and there is nothing for you to configure. A binary that predates the flag refuses to parse it; ChatCLI notices once and retries without it, so older installs keep working too. Set DEVIN_CLI_RESPECT_WORKSPACE_TRUST=true if you would rather see the failure than have the check waived.

​
Models

The account is the source of truth. /switch --model (and the completer, the @model tool and the ACP/MCP model pickers) asks the CLI itself — devin models list --format json — so the list is exactly what your account can invoke, Enterprise Team Settings restrictions included. Every entry the CLI reports is tagged [api]; the CLI-reported context window and output cap are registered on the fly, so a brand-new model or variant gets real budgets (compaction, ctx% footer, /max-tokens ceiling) instead of the generic fallback. The listing carries the family slugs the --model examples use (claude-opus-5, gpt-5.6-terra, swe-1.7-lightning) and their variants — reasoning levels and modifiers such as claude-opus-5-high, gpt-5-6-sol-max-priority, glm-5-2-1m, swe-1-6-fast — plus the CLI’s short aliases (opus, sonnet, codex, gemini, swe) resolving to the right family’s specs. Legacy enum-style ids the CLI still reports for a few older families (MODEL_…) are skipped; those families stay reachable through their slug. The static catalog is the fallback, not the gate: it covers the CLI being unreachable or not logged in (entries without the [api] tag), and any model string still passes through untouched.

​
How the transport works

  • Stateless per turn — the full flattened history goes on every call (never --resume), so conversation state never splits between ChatCLI and Devin’s servers. Compaction, /session load and context edits keep working unchanged.
  • The inner agent can’t act — each call runs in a fresh empty directory with a transport preamble that forbids Devin’s native tools while explicitly deferring to ChatCLI’s own textual tool protocol. In agent/coder modes the model sees ChatCLI’s full tool catalog and emits <tool_call .../> markup normally — ChatCLI executes, not Devin.
  • No identity coercion — the preamble never tells the model it “is” ChatCLI and demands no secrecy about the transport. The agent keeps its own identity (ask it who it is and it answers truthfully) and simply cooperates through the textual protocol. This matters in practice: an earlier preamble that assigned an identity plus a “don’t mention these rules” clause made Devin refuse whole tasks rather than misrepresent itself.
  • Lenient tag parsing — models backed by agent CLIs (Devin, Codex, Claude Code) sometimes shorten the <tool_call ...> tag to <tool ...>. ChatCLI’s parser accepts both spellings everywhere (agent, coder, chat exceptions, MoA, MCP server) while always emitting the canonical <tool_call> in its own prompts — liberal in what it accepts, conservative in what it sends.
  • Workspace trust is waived, deliberately — from the 3000.6 CLI line onward, --print refuses to run in a directory it has not been told to trust, and it has no way to raise the trust prompt in non-interactive mode. Every ChatCLI turn runs in a private os.MkdirTemp directory created moments earlier that holds nothing but the prompt file, so there is no project there to trust or distrust — ChatCLI passes --respect-workspace-trust false and the provider keeps answering. A binary that predates the flag refuses to parse it; that rejection is latched process-wide and the turn is retried without it, so older installs keep working. DEVIN_CLI_RESPECT_WORKSPACE_TRUST=true restores the CLI default, and pinning the flag yourself through DEVIN_CLI_EXTRA_ARGS suppresses ChatCLI’s copy — passing it twice is a parse error.
  • Clean replies — the answer is extracted between sentinel markers so Devin’s harness chrome is discarded; prompt files are always coerced to valid UTF-8 (the CLI rejects invalid bytes); invocations are serialized per process so background work (memory extraction) never races a live turn.

​
Environment variables

All exposed in /config providers.

​
Limitations (honest ones)

  • Cost is derived, not billed — every turn reads its real token usage back from the CLI’s ATIF export, and the rate is the one the CLI lists for your account (cost_summary), so /cost reports what the routed model costs at Cognition’s per-token rate. An enterprise build whose listing carries no cost_summary falls back to a static table of Cognition’s rates mirrored from an account listing, and CHATCLI_MODEL_PRICING (DEVIN:model=input/output;DEVIN:*=input/output, USD per MTok) outranks both when your contract differs. ACUs are not modeled; the pricing layer never bills a routed claude-*/gpt-* model as if it were the direct API, a family no listing ever priced is named by /cost as having no known rate rather than shown as $0, and a build without --export degrades to the chars/4 estimate.
  • Spawn latency — each turn pays a subprocess start plus Devin’s harness overhead. If that matters for heavy coder use, an ACP-based persistent transport (devin acp) is the natural evolution.
  • Vision doesn’t pass through — the transport is flat text; image attachments never reach the backing model.

​
See also