apunkt/plano
1
0
Fork 0
mirror of https://github.com/katanemo/plano.git synced 2026-06-14 15:15:15 +02:00
Code Issues Projects Releases Packages Wiki Activity

feat(claude-cli): add local Claude Code CLI provider bridge

Browse source 
Spawn the local `claude` binary as a subprocess and expose it as an
Anthropic Messages-compatible provider. Hosted in brightstaff
(`CLAUDE_CLI_LISTEN_ADDR`), with session reuse, idle TTL, and watchdog.

User-facing surface is `model_providers: [{ model: claude-cli/* }]` —
the Python CLI auto-fills name/provider_interface/base_url/access_key
and the launcher (native + supervisord) enables the bridge listener
only when at least one claude-cli provider is present.
This commit is contained in:
Spherrrical 2026-05-04 12:57:53 -07:00
parent b71a555f19
commit 9fdfeb7cbf
26 changed files with 2847 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

49
demos/integrations/claude_cli/README.md Normal file
View file

@ -0,0 +1,49 @@
# Claude Code CLI as a Plano provider
This demo wires the locally installed `claude` binary as a Plano
`model_provider`. The single line under `model_providers:`
```yaml
model_providers:
- model: claude-cli/*
default: true
```
is enough to:
1. Auto-fill `provider_interface: claude-cli`, `base_url: http://127.0.0.1:14001`
and a placeholder `access_key` (the CLI uses its own login keychain).
2. Start a localhost bridge inside `brightstaff` that spawns `claude -p
--output-format stream-json --input-format stream-json` for each
conversation.
3. Expose every Claude Code model — `claude-cli/sonnet`, `claude-cli/opus`,
`claude-cli/haiku`, plus dated full ids — at `GET /v1/models`.
## Running
```bash
# Make sure the CLI is logged in. You can use API krey billing or a paid Claude subscription.
claude auth login
# Start Plano in native mode.
planoai up demos/integrations/claude_cli/config.yaml
```
Then point any OpenAI- or Anthropic-style client at `http://localhost:12000`
and pick any `claude-cli/...` model. Plano routes the request through Envoy
to the brightstaff bridge, which asks the local `claude` binary to handle
it.
## Optional overrides
Set these env vars before `planoai up` if you need to tweak the bridge:
| Env var | Default | Meaning |
| ----------------------------- | ------------------- | -------------------------------------- |
| `CLAUDE_CLI_BIN` | `claude` | Path to the CLI binary. |
| `CLAUDE_CLI_PERMISSION_MODE` | `bypassPermissions` | `--permission-mode` flag value. |
| `CLAUDE_CLI_LISTEN_ADDR` | `127.0.0.1:14001` | Bridge listen address. |
| `CLAUDE_CLI_SESSION_TTL_SECS` | `600` | Idle TTL before a child is killed. |
| `CLAUDE_CLI_WATCHDOG_SECS` | `120` | Per-line watchdog inside one CLI turn. |
| `CLAUDE_CLI_MAX_SESSIONS` | `64` | Hard cap on concurrent CLI children. |

27
demos/integrations/claude_cli/config.yaml Normal file
View file

@ -0,0 +1,27 @@
version: v0.4.0
# Claude Code CLI as a Plano model_provider.
#
# The single line below is everything you need: Plano detects the
# `claude-cli/*` namespace, auto-fills the provider_interface, base_url and
# placeholder access_key, and starts a localhost bridge inside brightstaff
# that shells out to the `claude` binary on your $PATH for each request.
#
# Requirements:
# - `claude --version` must work in the same shell as `planoai up`.
# - Auth happens via the CLI's own `claude auth login` (no API key needed
# in Plano).
#
# Optional overrides via env (set before `planoai up`):
# CLAUDE_CLI_BIN=/custom/path/to/claude
# CLAUDE_CLI_PERMISSION_MODE=default
# CLAUDE_CLI_LISTEN_ADDR=127.0.0.1:14001
listeners:
- type: model
name: model_listener
port: 12000
model_providers:
- model: claude-cli/*
default: true